You are on page 1of 10

Challenges and Ways Forward for Avionics

Platforms and their Development in 2019


Bjoern Annighoefer Martin Halle Andreas Schweiger, Marina Reich
University of Stuttgart Hamburg University of Technology Airbus Defence and Space GmbH
Stuttgart, Germany Hamburg, Germany Manching, Germany
bjoern.annighoefer@ils.uni-stuttgart.de martin.halle@tuhh.de {andreas.schweiger,marina.reich}@airbus.com

Christopher Watkins Steven H. VanderLeest Stefan Harwarth, Patrick Deiber


Gulfstream Aerospace Corporation squishLogic LLC Wind River GmbH
Savannah, GA, USA Grand Rapids, MI, USA Ismaning, Germany
chris.watkins@gulfstream.com steve@squishlogic.com {patrick.deiber,stefan.harwarth}@windriver.com

Abstract—Today’s air vehicles depend on digital technology. The number of avionics research summaries is limited.
It accounts for more than 30% of their development costs. The Bieber et al. stated the integration of Commercial-off-the-
number of functions, the lines of code, the degree of autonomy, shelf (COTS) and multi-core processors into avionics as being
and the number of vehicles rise. This is why there is a need
for cutting-edge technology and development methods. There is most important and proposed a predictive execution model [2].
a gap between academia’s methods and industrial applications Goodloe et al. indicated significant challenges in safety and
due to multi-disciplinary challenges. We summarize the state- security in autonomy [3]. They proposed formal methods
of-the-art in avionics, namely avionics platforms, requirements and cyber-physical system models as ways forward. Gaska
engineering, model-based development, automated verification, et al. observed major challenges in the rising complexity
emerging technologies, and emerging demands. Experts review
the most demanding challenges, research gaps, and promising and development effort of avionics platforms [4]. They saw
solutions. They provide recommendations for the enhancement a solution in open architectures and an economic dual-use
of the cooperation between industry and academia and suggest between avionics, the automotive domain, and Internet-of-
necessary research topics. This article is an introduction for those things (IoT). Our contribution is to update the picture as
who are new to avionics. It is an up-to-date summary, for insiders of 2019, necessarily broadening the scope since there are
looking for most promising solutions to their current problems;
and it is a guide for those advancing avionics research. major multi-disciplinary challenges. This is also a result of the
Index Terms—avionics platforms, requirements engineering, AvioSE’19 workshop [5]. We target academia by quantifying
model-based development, automated verification, multi-core the current needs of the industry. We target industry by
summarizing innovative approaches and methods of academia.
I. I NTRODUCTION Topics dealt with are:
The term Avionics was coined in the 1960s. Today, avionics
• Avionics platforms, i.e. Integrated Modular Avionics,
accounts for over 30% of the aircraft development costs [1].
safety-critical COTS, middleware, and self-organization
Software (SW) and electronics provide a main unique selling
• Requirements engineering, i.e. Natural Language Pro-
point between aircraft and their airframers. There is a gap
cessing (NLP), Controlled Natural Language (CNL), and
between required cost and effort vs. acceptable ones and this
formal specification
gap enlarges. New programs, small electric aircraft, air taxis,
• Model-based development, i.e. general purpose or
drones, and autonomous vehicles are under development. The
domain-specific modeling languages, model transforma-
cost per vehicle must be low. Moreover, the targeted degree
tion, code/document generation, and early validation
of autonomy is high. The situation demands for cutting-edge
• Automated verification, i.e. virtual verification, theorem
methods of systems and SW engineering, i.e. formal methods,
proving, and model checking
model-based development, and development automation. For
• Emerging technologies, i.e. multi-core CPU, Time Sen-
instance, (1) the costs for the development of a digital fly-
sitive Networks (TSN), optical/power-line/data-centric
by-wire system are still almost the same as for the first one.
communication, virtualization, and containers
(2) In the future, only multi-core processors will be available,
• Emerging demands, i.e. security, artificial intelligence,
but certification remains quite challenging. (3) Autonomous
and pilot-reduced/less cockpit
vehicles demand functions of a complexity that can hardly be
certified with the current methods. New and more efficient For each topic experts review the current state-of-the-art
methods are under research or already available. However, and most important gaps are given in Sections II to VII. The
discrepancies exist between what is needed in industry and results are condensed in Sec. VIII to higher order research
what is developed in academia. recommendations and summarized in Sec. IX.

978-1-7281-0649-6/19/$31.00 ©2019 IEEE


Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
II. AVIONICS P LATFORMS every new aircraft presents significant differences in configura-
Avionics platforms are state-of-the-art, composed of stan- tion, HW compatibility, development environment, etc. Along
dardized computers, SW and bus systems. One party provides with the inevitable new features for a new product, adaptations
the platform, others provide the SW, e.g. system functions. of existing certification artifacts as well as development of
Platforms strive to reduce Size, Weight, and Power, as well as new artifacts are forced. Thus, finding a practical approach
Cost (SWaP-C). However, platforms also exhibit challenges in for reusable modular certification is a challenge that DO-297
integration, certification, performance, and safety. does not sufficiently address.

A. Integrated Modular Avionics C. Operating Systems and Middleware


Integrated Modular Avionics (IMA) are avionics platforms Operating Systems and Middleware (MW) increase reuse of
that share computing, Input/Output (IO) interfaces, and net- avionics SW, e.g., standardized MW for redundancy manage-
work between several aircraft functions. Its benefits lead to ment. [9] The general conflict between standardization with
virtually every airframer leveraging IMA [4] and achieving approved certification artifacts and customization is addressed
SWaP-C reductions. Moreover, IMA is the focus of important by initiatives such as Modular Open System Architectures
avionics standards such as DO-297 incremental certification; (MOSA) [10] and the FACE standard developed by a con-
ARINC 653 operating system API; and ARINC 664 rate- sortium of government, industry and academia, combining
constrained Ethernet-based bus. Space and time partitioning existing SW standards (ARINC653, POSIX, DDS) [11] for
enabled co-hosting mixed-criticality functions, where certifi- a segmented operating environment with interchangeable OS
cation artifacts can be reused hosting an unmodified software and MW components, similar to iOS or Android. Moreover,
module. However, applications remain incompatible between connectivity within aircraft systems and to the Internet is
IMA systems. Detailed APIs, usage-domain rules, and con- bringing cyber-security requirements into formerly isolated or
figuration parameters differ. Platform compatibility standards air-gap protected avionics systems. Continuous airworthiness
would ease reuse and integration. Moreover, when a system is requires permanent monitoring for security threats and their
architecturally mapped onto IMA, the inter-system dependen- safety impact, as discussed in Sec. VII-A and [12]. Also,
cies are not obvious, e.g. obscured by the network ”cloud”. the transition to multi-core processors and to architectures
Developers struggle with questions like ”Whom do I depend other than PowerPC creates challenges for OS and MW layers.
on?”, and ”What are the system boundaries?” since their Parallel execution of applications introduces interference that
function is spread across many distributed components [6]. complicates the demonstration of determinism and partitioning
These dependencies are obscured in configuration data spread as discussed in Sec. VI-A. A way forward are new processor
across thousands of files. Graphical system modeling tools are features such as IO Memory Management Unit (IOMMU),
needed to help system integrators perform integration, and performance counters, or cache partitioning that can be lever-
help system owners understand the integration. In addition, aged both for verification, and at run-time for partitioning or
machine-readable models could make the platform aware of fault containment [13]. Virtualization as detailed in Sec. VI-C
expected data-flows and could report missing/rogue data in can be used by OS and MW to enable rehosting of previously
situations where measurements are insufficient (e.g., wire developed applications on new architectures [14].
damage). SWaP-C could further be reduced with tighter inte-
D. Self-organizing Avionics
gration of IMA with display and power distribution functions.
Commonality in tools, IO models, and architectures should The human labor of configuration could be eliminated
be leveraged. Reconfiguration can reduce platform size and with self-organizing platforms. Self-organization means that
increase availability by reallocating resources temporarily. computers detect peripherals, organize work share, and ensure
Current approaches struggle with scenario explosion in cer- safety and security. Moreover, doing so during operation
tification [7]. Enabling certification frameworks and formal could increase safety margins by reconfiguration capabilities.
methods are missing. Technological feasibility was shown in a less-critical cabin
domain [15]. Safety-critical applications lack deterministic al-
B. Safety-critical Commercial-Off-The-Shelf (COTS) gorithms for self-organization and self-proofing. First attempts
Today, most platform modules are custom developed for a were made for topology verification [16]. However, self-
single platform. COTS hardware (HW) with associated certi- organization contradicts current approaches to certification. A
fication artifacts could improve portability and decrease cost, way forward could be that the platform mimics the behavior
allowing manufacturers to reuse technology developed within of a certification authority. Trust could be provided by the
one program in other domains, e.g. high power switching evidence that the platform can provide any certification item
and dual-lane IMA devices [8]. However, additional effort is in any state.
required since tool environments differ, which must be tailored
to specific HW features. For instance, HW or Operating III. R EQUIREMENTS E NGINEERING
System (OS) configuration tools are usually not generic. Requirements Engineering (RE) is the process of develop-
An ARINC 653 compliant COTS OS aims at providing ment and maintenance of requirements, and is expected in the
modular certification that can be reused. However, in practice, guidelines ARP-4574A for aircraft and system development,

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
DO-254 for HW, and DO-178C for SW. A major challenge e.g. by using logical expressions [21] and can even generate
is the assurance of requirement quality, i.e. identification of design models [22].
defects like incompleteness, inconsistency, and ambiguity, but Successful production tools are already available [23], but
also to identify errors like undefined conditions and missing a hindrance in their use is the significant effort creating glos-
quantification. saries, ontologies, and adapted patterns. There is no common
Development efficiency, analysis/review effort, and testing database of patterns or ontologies. Another weakness is that
are influenced by the quality of the requirements. Therefore, expressiveness is restricted by the finite set of patterns, i.e.
the following sections present current approaches in capturing remaining requirements might not be representable in the tool.
requirements that promise to ensure quality. Only approaches
that enable automation in RE are considered. C. Formal Specification
Formal specifications facilitate compliance between require-
A. Natural Language Processing (NLP) ments and the design. For the most of them, the created
Requirements in industry are mostly captured in text. Ex- specification can directly be utilized as inputs to formal
ample tools for requirements management include DOORSTM , verification techniques, model-checking, and theorem prov-
Jama ConnectTM , and Microsoft® Word. Natural Language ing (see Sec. V). With property-oriented techniques, such
Processing (NLP) is an approach for automated textual de- as Larch [24], OBJ [25], and Common Algebraic Specifi-
composition and analysis of man-made textual requirements. cation Language [26] the specification is expressed by log-
First successful applications utilizing NLP are available for ical statements with algebraic semantics, e.g. based on set
automated quality assurance of requirements [17], for in- theory. These specification languages are declarative formal
stance the assurance of INCOSE criteria and compliance to languages. Model-oriented techniques, such as Vienna Devel-
ISO/IEC/IEEE 29148:2011. NLP can also help classify and opment Method [27], Z [28], and B [29] provide the necessary
cluster requirements, extract key abstractions, generate models, language constructs to specify the behavior of an operation on
and traceability. Starting from these visions, a comprehensive an abstract, often declarative level, i.e. without stating how
summary of NLP challenges in RE were identified in the the system shall do it. A possibility is to state pre- and/or
NLP4RE 2018-workshop [18]: (1) Not enough input data is at post-conditions. Frequent constructs of processes in embedded
disposal for academic researchers to evolve single NLP tasks. systems are best captured with process algebra techniques, i.e.
Examples of requirements and their manual appraisal, e.g. the Calculus of Communicating Systems [30], e.g. Commu-
identification of defects or its classification, is required for the nicating Sequential Processes [31] and π-calculus [32]. With
training of NLP techniques. (2) NLP experts, RE researchers, Promela [33] it is possible to specify sequential and concurrent
tool vendors, and industry representatives do not sufficiently programs.
cooperate. Tailoring of NLP to the needs of the user, for The main hindrance is the formalization of requirements in
instance by compensating a small number of provided data by formal languages, that is in itself prone to errors and requires
incorporating domain-specific conditions, could reduce com- specific education of the engineer.
plexity in NLP applications for specific industries. (3) Domain-
specificity is not sufficiently considered in NLP techniques. D. General Requirements Engineering Challenges
Understanding of the mutual effects between diagrams, figures, Besides missing domain-specificity of most considered tech-
and textual requirements can contribute to capturing implicit niques, all approaches lack a connection to avionics-specific
knowledge. Moreover, domain-specific ontologies can bring physical models often created during the development of the
great value to the extension of the natural language considered system’s functionality. The missing measurement for overall
in public NLP. requirements quality was identified as a further root cause
Joint project reports by industry and academia are available for challenges during the AvioSE’19 workshop [5]. These
on the interplay of NLP and RE. The research efforts con- measurements exist for some of the above mentioned quality
centrate on enhancement of NLP for quality assurance [17], criteria, but are in general considered incomplete. During the
such as detection of ambiguity, comparatives, loopholes, neg- mentioned workshop also the lack of knowledge in RE was put
ative terms, non-verifiable terms, pronouns, subjectivity, and in focus. Finally, current empirical studies into waste due to
superlatives. However, these techniques lack in exploiting inefficient RE could boost acceptance for future improvements,
domain knowledge. Tailored parsing rules incorporating defini- by providing reliable estimates of current wasted human
tion of domain-specific glossaries and ontologies need further resources or numbers of defects traced back to RE.
advances.
IV. M ODEL - BASED D EVELOPMENT
B. Controlled Natural Language (CNL) Model-based based systems engineering (MBSE) is the
CNL goes beyond quality assurance of requirements as utilization of models in systems development. Objectives are
this technique applies certain rules for textual requirements: to speed up development steps, encapsulate complexity and
grammar, syntax, semantics, and structure. Besides quality increase design maturity via early functional validation. In
assurance by suppressing adverse terms and expressions [19], general, models provide a graphical representation, which
[20], CNL can enable the deduction of formal specifications can be understood naturally, are more rigid than text, and

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
validatable through use of model testing. In model-driven understanding often difficult. Graphical MTL are more in-
engineering model formats are defined and machinable. In tuitive to read. They describe transformations in diagrams
model-integrated computing SW is completely composed au- similar to the model itself, e.g. GREAT and Eclipse Henshin,
tomatically from models [34]. The following is a summary of but their development is not finished. The major challenge
MBSE applications in avionics. is to convince certification authorities to allow credit for
transformation methods. There is no formally proven model-
A. General Purpose Modeling Languages transformation language.
There are two categories of modeling: (1) Architecture
models depict structure, behavior, and system integration, D. Code and Qualification Artifact Generation
e.g. at the IO level. Model languages originate from SW The earliest application of model-driven engineering in
development, mostly based on the Unified Modeling Language avionics is the generation of SW. It was first matured by
(UML), which describes class or component dependencies, ANSYS SCADE, which generates qualified source code and
as well as discrete or logical behavior. A systems engi- documents. Generation is, however, restricted to SW interfaces
neering extension is SysML. Majorly used in avionics are with no physical interaction and no concurrency. Both exist
class diagrams and state machines. In addition, more specific in reality, thereby causing additional effort in development,
languages were derived for avionics, e.g. the Architecture testing, and qualification by wrapper code and configuration.
Analysis and Design Language (AADL), ARCADIA, and A zero-effort model-to-product generation seems impossible,
System Composer. AADL is a general system architecture because HW dependencies cannot be considered in necessary
language providing avionics extensions [35]–[37]. Arcadia was detail in any physical model today. If, however, the HW is
developed as alternative to SysML, implemented by Thales restricted and a unifying SW layer is introduced that guarantee
in the open source tool Capella. In March 2019, Mathworks a defined behavior, an almost complete qualifiable generation
released System Composer, based on their existing Simulink is possible. This is demonstrated with the Flexible Avionics
model constructs. (2) Physical models represent time-related Platform and the AAA process [43]. Implementations, tests,
physical behavior. Physical models are common for aircraft and qualification documents are generated with model trans-
systems, but not for avionics. formations. In this case, generation is not qualified, because
the resulting documents are equal to a human development
B. Domain-specific Modeling process. Two CS-23 fly-by-wire systems were developed with
Domain-Specific Modeling (DSM) is the creation of cus- AAA [44], [45]. Acceptance lacks a completed qualification
tomized, formally defined, model formats. A Meta-Modeling process. This work is ongoing. A lesson learned is that
Language (MML) defines the structure of a Domain-Specific generated documents tend to be more detailed and difficult
Language (DSL). The formal nature inherently allows ma- to read than documents created by humans.
chine processing and validation. DSM is extensively applied
in aerospace for customized structural models that simplify, E. Early Validation
streamline, and automate development steps, e.g. configuration Validating system functions on avionics platforms happens
management [38], avionics platform design, and optimization late although SW is developed in parallel to HW, because
([39], [40]), and development data management [41], resulting final testing depends on the HW being physically available.
in significant time savings, fewer errors, and improved under- Moreover, the increasing number of system functions and their
standing. With Meta Object Facility (MOF) there is a well- interoperability account for hidden failures due to the coupling
defined meta-modeling standard. Frameworks like Eclipse with the avionics platform, which is hard to assess and often
Modeling Framework (EMF) and Generic Modeling Envi- result in gaps in the specification. Methods for early validation
ronment (GME) are well established. Those tools, however, are desired that simulate abstract platform and system models
are generally kept out of the qualification path, because their to test functional and timing behavior in a virtual integration
impact on qualification is unclear and the frameworks (and based on operational scenarios. MBSE provides means to
all derived tools) are usually considered unqualifiable. Many validate model functionality prior to generating the code. A
originate from non-profit non-avionics communities. challenge is achieving the necessary level of detail of the
avionics platform model [36], I/O [46], the system functions,
C. Model Transformation and the emulated equipment. Furthermore, the number of inter-
DSLs of avionics systems allow for automated transforma- connections, states, and failure modes increases exponentially,
tion between models. For instance, an avionics architecture the more system functions are taken into account. Efficient and
model from pre-design can be converted to an Interface automated simulation generation, parametrization, testing, test
Control Document (ICD), application code, and configuration script and documentation generation (e.g. using model trans-
data [42]. Model Transformation Languages (MTL) describe formation techniques) are needed. Blind spots can be identified
transformations formally. Examples are XSL Transformation using scenario-based testing in addition to requirements based
and Query View Transformation. An efficient use, however, testing. Modeling standards and tool chains are required that
requires all models to be defined with the same MML and support early validation methods. First steps are presented
framework. Moreover, the level of abstraction is high and in [47].

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
V. AUTOMATED V ERIFICATION tation extends established ones [53] and can also be transferred
Automated verification aims at machine-based detection of to e.g. sequence diagrams [54].
FOCUS and its accompanying modeling tool Auto FOCUS
defects due to specification or implementation errors through-
out the development process. Approaches applied in avionics 3 [55] has seen a wide acceptance in academia [56]. In
are described in this section. particular, Kriebel et al. [56] present an approach for trans-
lating FOCUS models created in a tool [57] to an Isabelle
A. Virtual Verification automaton specification, which in turn is mapped to a set of
functions processing streams. This transformation allows for
Due to tighter development cycles, an improvement in
the verification of properties by the application of the theorem
the identification and correction of defects is an ever-lasting
prover Isabelle [58]. Nevertheless, there are still some issues to
desire. The earlier defects are identified in the development
be dealt with, before this approach fits for a wide application
process, the less expensive is their correction [48]. However,
in industry [56]:
dynamic test methods are only applicable after the complete
implementation of the SW. In other words, these tests are • Improved tool support
conducted rather late in the development process. • The robustness and level of automation of the proof
Schamai et al. [49] present an approach, which allows functionality have to be increased. In particular, the
for the verification of the system design against the system formal verification should be hidden completely from
requirements, i.e. at the very beginning of the development the developer. This in turn allows for a continuous
process. They suggest the following steps: verification, as the underlying idea behind this approach
is already state-of-the-art for continuous integration [59]
• Requirements selection: Out of the available textual re-
and delivery [60].
quirements those are selected manually that are amenable
for formalization. Though Isabelle has already been used successfully in
• Requirements formalization: Properties to be checked proving the correctness of a micro-kernel scheduler [61],
automatically are extracted manually from the selected the correctness of the proof is dependent on the correctness
requirements. Furthermore, corresponding models for of Isabelle and its implementation. Kunčar [62] proved the
monitors1 are created. completeness and correctness of a part of Isabelle’s code.
• Design model creation and selection: Particular design However, there are still Isabelle elements to be proved.
artifacts2 , which correlate with the selected requirements, Hoare’s calculus [63] provides the foundation for tool
are created and chosen. implementations supporting programming languages such as
• Linkage between requirements and design model prop- C (Frama-C, [64]) or Ada (SPARK, [65]). Though such tools
erties: For connecting the properties of both the require- have already been deployed successfully in industry [66]
ments and the design models, appropriate test models are according to DO-178C and its supplement DO-333, theorem
created. provers need the (possibly error-prone) user-interaction with
• Observation of potential requirement violations: While the provision of additional input in the form of lemmas
the created models are executed, potential violations are and also a highly-elaborated specification of properties to be
identified, and reported by monitors. verified.
While the desire to verify artifacts in the development
C. Model Checking
process as early as possible is valid, the described approach
involves a considerable amount of additional manual work. Model checking [67] is defined as an automatic method
Since in particular the description of properties is not a trivial for the verification of a model with respect to a specification
task, a lot more automation is necessary. In addition to this, the expressed in a mathematically-sound language. Since the com-
monitor specification is still a considerable challenge. The au- plexity of software is rising continuously, the state space of
tomatic deduction of such monitors from formal requirements both the model and its specification is increasing as well. This
could alleviate their application in industry. state explosion reduces the scalability of this approach.
In order to get around this issue, Esparza and Heljanko [68]
B. Theorem Proving use partially ordered graphs instead of deploying state transi-
The specification of software functionality can be conducted tion systems. This unfolding approach has seen considerable
using formal languages like FOCUS [52] with its well-defined improvements after its original presentation [69]–[71], which
and executable semantics. This language describes compo- checked only the reachability of states and identified possibly
nents with hierarchical structure using several semantically- existing deadlocks. Now unfoldings are e.g. amenable to
equivalent notations. In particular, the proposed graphical no- properties expressed in Linear Temporal Logic and supported
by various tools (see e.g. [72]).
1 A monitor (or observer) checks the potential violation of the modeled
Though the presented approach is very promising, its suit-
properties. ability for an application in industry needs to be confirmed. In
2 The model is designed using the language ModelicaML [50]. ModelicaML
enriches the modeling language Modelica [51] with UML’s graphical notation particular, the time and space efficiency of the implemented
by providing a UML profile. verification algorithms has still to be demonstrated [68].

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
VI. E MERGING T ECHNOLOGIES C. Virtualization and Containers
Avionics depends on HW of the general information tech- Partitioning is expected by IMA and required by ARINC
nology domain. The increasing technological progress induces 653. Virtualization abstracts computing HW so that it ap-
challenges especially in processors, communications, OS, and pears as a virtual resource. A hypervisor hosts multiple guest
displays. operating systems (GOS), innately providing partitioning by
managing the computing HW in lieu of the hosted software.
A. Multi-core CPU Virtual vessels called domains or partitions are serviced by
Moore’s Law and a frequency limit at approximately 4 a virtual machine running on one or more processor cores.
Ghz result in an increasing number of cores per processor. Separation is enforced with a timer that can interrupt the
What scales well in general IT “complicates control and adds current thread of execution and a Memory Management Unit
significant challenge to assurance analysis, because multiple (MMU) that segregates pages of memory. Thus the entire GOS
partitions can now run simultaneously on different cores and is partitioned with respect to time and space. In some cases
thus might interfere with one another more directly than separation of IO is also done by an IOMMU. These HW mech-
on single-core systems, where partitions must run one at a anisms are guarded from GOS manipulation via hierarchical
time” [73] in avionics. The key challenge is bounding the protection rings. Since virtualization provides time and space
Worst Case Execution Time (WCET), particularly cross-core partitioning implicitly, it is a natural fit for the partitioning
interference for memory resources and data paths e.g. for PCIe needs of avionics platforms. Moreover, certification is simpli-
and CPU interconnects. Unbounded interference jeopardizes fied since file systems, network protocol stacks, and graphical
the ability to provide a consolidated platform as safe as user interfaces are untouched as they are part of the GOSs. On
equivalent federated systems [74]. Additional challenges come the other hand, the specific Application Programming Interface
with hidden, or non-static microcode. The position paper (API), communication mechanisms, and health monitoring
CAST-32A makes remarks on the certification for multi-core dictated by ARINC 653 must be developed or added [77].
processors, but there are currently no certification guidelines In addition, virtualization adds a small performance overhead.
nor common agreements about necessary design steps and the Current measures state 1 to 20% [78], but this does not cover
dependability level of tests. Simplifications can be achieved the safety-critical domain and needs further investigation. For
with fully open processor architectures (e.g. ARM, RISC-V) example, determinism and non-interference properties required
or safety nets. for safety might require expensive cache flushing during a
partition switch. Performance issues can be addressed with
B. Advanced network technology
containers (e.g. Docker). Containers partition at the OS level
Substituting proprietary and single-purpose avionics busses and do not require HW virtualization extensions. They are
with open, general-purpose COTS communication systems light-weight in terms of source code size and performance
can greatly reduce SWaP-C. Historically, COTS networks impact. However, isolation is harder to assure without the
(e.g. Ethernet) lacked predictability and determinism. Special HW enforced separation of GOS. Currently, no approach is
solutions as ARINC 664 – a rate-constrained avionics Ethernet known using containers to isolate mixed-criticality software in
– were developed, which are very successful, but also costly an avionics system. However, a number of research projects
due to the limited market. A transition to COTS Ethernet might test containers to enable more rapid development, e.g. [79].
be feasible with the emerging Time Sensitive Networking As hypervisor performance continues to improve, a hypervisor
(TSN, IEEE 802.1Q-draft) standard. TSN standardizes time- based container approach such as KATA might be more
scheduled traffic and provides time-synchronization (IEEE appropriate for avionics.
802.1AS-rev) like TTP and TTE. TSN is superior to A664
in determinism and performance, while using low-cost COTS D. Glass cockpit
components. Open issues are, however, HW certification, Over the past few decades, the glass cockpit has evolved
redundancy, and a significantly higher configuration effort for from separate displays, to a combined display system con-
time-synchronized networks. Besides protocols, novel physical sisting of a suite of heads-up, heads-down, and smaller an-
and application layers are under consideration in avionics. cillary displays shared between system functions. Currently,
Power-line communication introduces new possibilities to save touchscreen technology is transitioning into the cockpit [80].
weight. However, taking safety (redundancy) or security into Moreover, tablets became a standard flight aid in the cockpit.
account, it makes the design harder by introducing addi- Novel approaches add touch to the fixed displays, such that
tional configuration parameters [75]. Optical transportation physical controls can be eliminated and replaced with soft on-
is immune to interference and lightweight, but lacks active screen controls. The latter can be replicated in many places.
switching capabilities, long-term experience, and configurabil- However, care must be taken as this flexibility can lend diffi-
ity. Network configuration can be simplified with data-centric culty in remembering control location. Moreover, existing user
solutions such as publish-subscribe or message queues. A interfaces are not optimal for touch/gesture controls and must
certified solution is a variant of DDS [76], which, however, be redesigned. The physical position of displays may need to
currently lacks a wide application, perhaps due to obscuring be closer to the crew and angled to reduce crew fatigue. As
configuration parameters within application software. display content increases, curved screen technologies should

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
be investigated to extend the display area in the inherently spend most of their time dealing with mission success. This
curved flight deck environment. The future is intriguing: shift has been accomplished by increasing the degree of
Wearable displays with augmented reality could provide a automation and carefree handling of the aircraft. Due to rising
tremendous field of view not attainable in traditional heads- mission complexity the workload will continue to increase in
up displays. Transparent displays are entering the consumer future, driving automation of the execution of missions as well.
electronics industry and could eventually be integrated with The incorporation of artificial intelligence (AI) could help.
aircraft windows to provide augmented visuals. The same applies for the development of avionics systems,
where engineers could benefit from automation through AI
VII. E MERGING D EMANDS algorithms. First approaches can be seen in NLP, but validation
New usage scenarios, as well as workload and cost re- and verification could be additional topics.
duction demand new avionics capabilities. Most challenging Machine learning as a subset of AI comprises simple neural
seem security, artificial intelligence (AI) integration, and crew networks and deep neural networks [86]. While simple neural
reduction. networks contain only one layer between input and output,
deep neural networks [87] consist of several hidden layers. The
A. Security
latter have already been deployed successfully in a setting with
While safety addresses faults, security prevents malicious considerable complexity [88]. Substantial progress has also
attacks. While safety assessment is routine in aerospace, secu- been reached in the context of safety-critical applications [89],
rity struggles mainly with lack of quantification, standardized resilience [90], and dependability [91] of neural networks.
representations, overall system knowledge, and hidden inter- However, it can still be demonstrated that adversarial exam-
domain links [81]. In a case study, Teso [82] highlighted ples [92] have negative impact on the dependability of this
several security vulnerabilities in the Aircraft Communications technology.
Addressing and Reporting System and the flight control sys- Though AI is already applied in non-safety-critical areas,
tem, e.g. message jamming and eavesdropping. Availability, major advancements are yet necessary. Up to now verification
confidentiality, integrity, and authentication are not imple- is impossible to prove, due to the difficulty of determining why
mented in the communication system [83]. Rising connectivity a given neural network has come to a certain decision. Having
emphasizes the above issues. a way to demonstrate this is a matter of uttermost importance
A semi-formal notation for security is UMLsec [84], which for the certification of a product in the avionics domain.
is a UML profile for the development and analysis of security-
critical systems. UML is extended with annotations for the C. Pilot-reduced/less Cockpit
specification of security requirements. The integrity of the Reducing the number of pilots in the cockpit is a logical
designed system model is ensured, if no constraints in the consequence of reducing operation costs. Single, remote or no-
UMLsec model are violated during simulated attacks. How- pilot operation have organizational, social, and technical chal-
ever, there is currently no means for confirming, whether lenges ([93], [94]). From the avionics perspective the general
the set of simulated attacks is sufficiently complete. This issue is: today pilots are the final authority in critical situations.
remains a key challenge with security, since testing against Removing even one pilot changes the situation: A healthy
known attacks is not a good indicator of robustness against pilot has an incapacitation probability of 10−6 1/h [95].
future (as yet unknown) attacks. While UMLsec is a passive This is insufficient. Therefore, the avionics system must be
countermeasure, Crane [85] suggest an active approach, where able to take over the final decision and new systems or
booby trapping inserts countermeasures automatically into the the combination of avionics and a pilot must have a failure
software. This approach is feasible for desktop applications, rate less than 10−9 1/h. Three significant avionics platform
but the following issues prevent deployment in safety-critical challenges are:
real-time software: • In single pilot operation a reliable redundant system
• Missing functional and temporal equivalence of the soft- composed of human and avionics can only be achieved,
ware with and without booby traps if the pilot’s integrity is known by the avionics through
• Missing deterministic and reliable identification of mali- reliable incapacitation detection devices.
cious attacks • In remote pilot operation requires a low-latency, high-
In general, formally proven software (see Sec. V) has some integrity, high-availability (10−9 1/h) secured global
promise for guaranteeing reliability even for yet unknown coverage civil aerospace data link. Neither a suitable
cyber-security attacks, because formal methods can mathemat- technology nor a global platform are hardly available
ically prove immunity to entire classes of attacks, not just today.
specific instances of test cases. However, the high cost of • Fully autonomous operation requires new autonomy func-
formal methods is still prohibitive. tions, such as air traffic control negotiation, system su-
pervision, minimal equipment list management, fuel and
B. Artificial Intelligence flight-plan supervision, recovery, and emergency plan-
While the main workload of an aircrew was dedicated to ning [96]. Autonomy functions will be more complex and
flying the aircraft one or two decades ago, nowadays pilots require significantly more computational power. This will

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
emphasize today’s lack in complex (AI) function and high investigated what is the best combination of methods that
performance hardware (multi-core) certification. could cover most development aspects. Moreover, standardized
In any case a more sophisticated interaction between pilots method interfaces should be defined, covering requirements,
and automation is required. This includes visualization tech- models, transformations, and equations and allow for a seam-
nology, control interfaces, and self-explanation of algorithmic less integration.
decisions. It must be clear: Who is in control, why, and what (5) Another significant hurdle in certification is HW/SW
they are doing? Moreover, this human-machine interaction dependency. System tests have to be carried out on real
must be an integral part of a suitable qualification process. hardware. Virtual testing is not possible due to insufficient and
non-qualified simulations, and reuse in terms of qualification
VIII. D ISCUSSION AND WAYS F ORWARD is very restricted due to missing standards at the HW/SW
All aspects of avionics exhibit individual challenges. Nev- interface. → Research should be carried out on reliable HW
ertheless, there are some general issues, which are suggested simulation technology as well as on a resilient abstraction of
to be addressed as follows: HW behavior.
(1) Many challenges are related to complexity. Avionics (6) New technologies evolving in processing, graphics,
platforms, SW, and HW become too complex to be fully networks, new vehicles, and vehicle automation do not bring
gathered by a human’s mind, such that it is a real effort very different challenges, but will exponentially emphasize the
to derive complete verification tests and documents as it is existing ones. Moreover, there is a gap between progress of
desired by the certification processes. Moreover, the number of general IT and its adaptation in avionics that could increase up
stakeholders in the development and distribution of knowledge to a point where there is no overlap anymore. The aerospace
rises. This can be addressed in two ways. First, by making it industry is too small to develop their HW completely alone
manageable in suitable models and, second, by the automation and neither customers, pilots nor passengers would understand
of development steps. Domain-specific languages and trans- that. → Effort in teaming up with other safety-critical cyber-
formations are good tools for managing complexity. Their physical system domains seems inevitable. The point in time
applications, however, lack mature tools and visualization of is almost perfect, since autonomous car developers begin to
large structural model data. → Effort should be spent in further realize that they have to go for an aircraft-like certification
developing modeling tools and methods such that they are process, but for their SW and HW this seems almost infeasible.
applicable to avionics development processes and could even Likewise for UAV, IoT, space, rail, and nautical domains. How-
be used in certified tools or SW. The efficient and reliable ever, this must be fostered by appropriate research projects and
visualization of complex system structures becomes inevitable a willingness from all domains to develop common results and
in future but must be advanced. to define and use domain-wide standards.
(2) Automation is provided by NLP, transformations, and
formal methods. All, however, lack trust. Humans are natu- IX. C ONCLUSION
rally skeptical of automated results and the tools generating
those results and automation algorithms are bad in explaining Avionics are mandatory in current flying vehicles and their
results. → Effort must be spent in finding out what evidences importance and performance expectations will rise in future.
automation should produce in order to make it trust-worthy Engineers already struggle with the complexity of systems,
and eligible for certification credits. SW, HW, and a heavily manual certification process. Addi-
(3) Trust is also a matter of education. IT is important in tional hot topics are the inevitable change to multi-core proces-
avionics, yet the majority of aircraft engineers are (and want sors, high-bandwidth busses, active security, and autonomous
to be) trained as mechanical engineers. Computer scientists on transportation. Ways forward are modeling, development au-
the other hand have less understanding in avionics develop- tomation, and the adaptation of IT standards, virtualization,
ment, qualification, and certification. Although they might be TSN, and touch inputs. Methods that allow the efficient
aware of methods that help, they do not know how the avionics development of future autonomous systems are available with
domain could benefit most from these methods. → Knowledge formal and computer-aided requirements engineering, domain-
on modeling, transformation, formal checkers, and security specific modeling, graphical model transformations, theorem
must be emphasized in education, such that young engineers proving, simulation, and test/code/document generation. How-
(and certification agents) know the expectable results and ever, there is a significant lack of trust, integration, compat-
drawbacks of a broad range of useful methods. Since it is ibility, education, and implementation of such methods that
surely not the only skill needed in avionics, an option might be prohibit a straightforward application today. It is suggested to
that the general aerospace student can opt for a specialization foster the research and academic and industrial cooperation
in ”digital systems engineering”. in these areas; namely define a method combination, create
(4) The broad range of methods is another field for improve- qualifieable implementations, and enshrine them in aerospace
ment. There is no one-size-fits-all method, but a combination education. It seems inevitable, moreover, to team up with other
can cover a significant portion of current issues. However, domains to achieve a critical mass in method and hardware
most methods are developed and advertised in their community development. Any progress in these areas is encouraged and
and interfaces between methods are bad. → It should be the authors are open for cooperation.

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
Future updates of this article are anticipated. The idea [22] L. Lúcio, S. Rahman, S. bin Abid, and A. Mavin, “EARS-CTRL:
is to take it as a baseline and document progress as it is Generating controllers for dummies,” in MODELS, 2017.
[23] The Reuse Company. (2019) Systems engineering suite.
made in the individual areas. Moreover, avionics are broader [24] J. V. Guttag and J. J. Horning, Larch: Languages and Tools for Formal
than discussed in this article. Areas, which must be covered Specification. Berlin, Heidelberg: Springer-Verlag, 1993.
in future include cabin electronics and secure programming [25] J. Goguen, C. Kirchner, H. Kirchner, A. Mégrelis, J. Meseguer, and
T. Winkler, “An introduction to obj 3,” in Conditional Term Rewriting
languages. Systems, S. Kaplan and J. P. Jouannaud, Eds. Berlin, Heidelberg:
Springer Berlin Heidelberg, 1988, pp. 258–263.
ACKNOWLEDGMENT [26] E. Astesiano, M. Bidoit, H. Kirchner, B. Krieg-Brueckner, P. D. Mosses,
D. Sannella, and A. Tarlecki, “Casl: the common algebraic specification
We thank Andy Bellis from Collins Aerospace for his language,” Theoretical Computer Science, vol. 286, no. 2, pp. 153 –
contributions and review of this article. 196, 2002, current trends in Algebraic Development Techniques.
[27] I. J. . Joint Technical Committee, Information technology — Program-
R EFERENCES ming languages, their environments and system software interfaces —
Vienna Development Method — Specification Language — Part 1: Base
[1] R. Collinson, Intro. to Avionics Systems, 2nd ed. Springer, 2011. language, ISO ISO/IEC 13 817-1, 1996.
[2] P. Bieber, F. Boniol, M. Boyer, E. Noulard, and C. Pagetti, “New chal- [28] S. S. . Joint Technical Committee ISO/IEC JTC 1, Information tech-
lenges for future avionic architectures,” AerospaceLab, vol. 4, no. 11, nology, Information technology — Z formal specification notation —
May 2012. Syntax, type system and semantics, ISO ISO/IEC 13 568:2002, 1996.
[3] A. Goodloe, “Challenges in the verification of flight-critical systems,” [29] J. Abrial and J. Abrial, The B-Book: Assigning Programs to Meanings.
in CPS V&V I&F Workshop, Pittsburgh, USA, 2014. Cambridge University Press, 2005.
[4] T. Gaska, C. Watkins, and Y. Chen, “Integrated modular avionics - past, [30] R. Milner, A Calculus of Communicating Systems. Berlin, Heidelberg:
present, and future,” IEEE Aerospace and Electronic Systems Magazine, Springer-Verlag, 1982.
vol. 30, no. 9, pp. 12–23, Sept 2015. [31] C. A. R. Hoare, Communicating Sequential Processes. New York,
[5] B. Annighoefer, A. Schweiger, and M. Reich, Eds., 1st Workshop on NY: Springer New York, 2002, pp. 413–443. [Online]. Available:
Avionics Systems and Software Engineering, 2019. https://doi.org/10.1007/978-1-4757-3472-0 16
[6] C. B. Watkins and T. Burns, “Evolution of the systems integrator role and [32] R. Milner, Communicating and Mobile Systems: The Pi-calculus. New
change management process within highly integrated aircraft systems,” York, NY, USA: Cambridge University Press, 1999.
in 2015 IEEE/AIAA 34th Digital Avionics Systems Conference, Sept [33] M. Ben-Ari, Principles of the Spin Model Checker, 1st ed.
2015. [34] J. Sztipanovits and G. Karsai, “Model-integrated computing,” Computer,
[7] P. Bieber et al., “Preliminary design of future reconfigurable IMA vol. 30, no. 4, pp. 110–111, April 1997.
platforms,” SIGBED Rev., vol. 6, no. 3, pp. 1–5, 2009. [35] J. Delange, L. Pautet, A. Plantec, M. Kerboeuf, F. Singhoff, and
[8] B. Kornek-Percin, “New IMA architecture approach based on ima F. Kordon, “Validate, simulate, and implement ARINC653 systems using
ressources,” in 2015 IEEE/AIAA 34th Digital Avionics Systems Con- the AADL,” in Proc. of the ACM SIGAda annual Intl. conference on
ference, Sept 2015, pp. 1–19. Ada and related technologies, ser. SIGAda ’09. New York, NY, USA:
[9] B. Luettig et al., “A service provisioning layer enabling simplex-minded ACM, 2009, pp. 31–44.
function development on integrated modular avionics hardware,” in IEEE [36] M. Lafaye, M. Gatti, D. Faura, and L. Pautet, “Model driven early
37th Digital Avionics Systems Conference, 09 2018. exploration of IMA execution platform,” in Digital Avionics Systems
[10] J. Tokar, “A comparison of avionics open system architectures,” ACM Conference, 2011 IEEE/AIAA 30th, oct. 2011, pp. 7A2–1 –7A2–11.
SIGAda Ada Letters, vol. 36, pp. 22–26, 05 2017. [37] Z. Li, S. Wang, and T. Zhao, “A model based simulation verification
[11] M. Garcı́a-Valls et al., “Using DDS middleware in distributed partitioned method for IMA reconfiguration on system level,” in 2015 First Intl.
systems,” ACM SIGBED Review, vol. 14, pp. 14–20, 01 2018. Conference on Reliability Systems Engineering, Oct 2015, pp. 1–5.
[12] A. Baker and P. Parkinson, “Cyber security enhancements for a safety- [38] M. Halle and F. Thielecke, “Next generation IMA configuration engi-
critical arinc 653 avionics platform,” in Twenty-sixth Safety-critical neering - from architecture to application,” in 2015 IEEE/AIAA 34th
Systems Symposium, York, UK, 2 2018. Digital Avionics Systems Conference, Sept 2015, pp. 6B2–1–6B2–13.
[13] L. H. Mutuel et al., “Assurance of multicore processors in airborne [39] B. Annighoefer and F. Thielecke, “A systems architecting framework
systems,” FAA, Tech. Rep. DOT/FAA/TC-16/51, 2017. for distributed integrated modular avionics,” in Deutscher Luft- und
[14] D. Radack, H. G. Tiedeman Jr, and P. Parkinson, “Civil certification of Raumfahrtkongress, Augsburg 16. - 18. Sept. 2014. Augsburg: Deutsche
multi-core processing systems in commercial avionics,” in 27th Safety- Gesellschaft fuer Luft- und Raumfahrt, September 2014.
critical Systems Symposium Proceedings, 02 2019. [40] L. Batista and O. Hammami, “Capella based system engineering mod-
[15] B. Annighoefer, M. Riedlinger, O. Marquardt, R. Ahmadi, B. Schulz, elling and multi-objective optimization of avionics systems,” in 2016
M. Brunner, and R. Reichel, “The adaptive avionics platform,” IEEE IEEE Intl. Symposium on Systems Engineering, Oct 2016, pp. 1–8.
Aircraft Electronics Systems Magazine, 2019. [41] J. Yin, B. Lawler, and H. Jin, “Application of model based system
[16] B. Schulz and B. Annighoefer, “A verification algorithm for the engineering to IMA development activities,” in 2017 IEEE/AIAA 36th
automatic topology discovery of the adaptive avionics platform,” in Digital Avionics Systems Conference, Sept 2017, pp. 1–7.
IEEE/AIAA 37th Digital Avionics Systems Conference, London, UK, [42] M. Halle and F. Thielecke, “Model-based transition of IMA architecture
Sep. 2018. into configuration data,” in 35th Digital Avionics System Conference,
[17] H. Femmer, D. Méndez Fernández, S. Wagner, and S. Eder, “Rapid Sacramento,CA , USA, September 2016.
quality assurance with requirements smells,” Journal of Systems and [43] T. Belschner et al., “Automated generation of certification relevant
Software, vol. 123, 03 2016. documentation for a distributed avionics platform approach,” in 2016
[18] F. Dalpiaz, A. Ferrari, X. Franch, and C. Palomares, “Natural language IEEE/AIAA 35th Digital Avionics Systems Conference, Sept 2016, pp.
processing for requirements engineering: The best is yet to come,” IEEE 1–10.
Software, vol. 35, no. 5, pp. 115–119, Sep. 2018. [44] Dalldorff, Luckner, and Reichel, “Full-authority automatic flight control
[19] C. Denger, D. M. Berry, and E. Kamsties, “Higher quality requirements system for the civil airborne utility platform S15 - LAPAZ,” in IEEE
specifications through natural language patterns,” in Proc. 2003 Sympo- 2nd CEAS Specialist Conference on Guidance, Navigation & Control,
sium on Security and Privacy, Nov 2003, pp. 80–90. 2013.
[20] G. Génova, J. M. Fuentes, J. Llorens, O. Hurtado, and V. Moreno, “A [45] Dries, W. Fichter, A. Joos, Mayerbuch, Müller, Pinchetti, R. Reichel,
framework to measure and improve the quality of textual requirements,” R.-R. Riebeling, Stephan, and Volck, “Automatischer start und land-
Requirements Engineering, vol. 18, no. 1, pp. 25–41, Mar 2013. ing einer diamond DA42,” in Deutscher Luft- und Raumfahrtkongress
[21] M. Autili, L. Grunske, M. Lumpe, P. Pelliccione, and A. Tang, “Aligning (DGLR), 2016.
qualitative, real-time, and probabilistic property specification patterns [46] B. Annighöfer, H. Höber, and F. Thielecke, “An easy-to-use real-
using a structured english grammar,” IEEE Transactions on Software time AFDX simulation framework,” in 35th Digital Avionics System
Engineering, vol. 41, no. 7, pp. 620–638, July 2015. Conference, Sacramento, CA , USA, 9 2016.

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.
[47] M. Halle and F. Thielecke, “Tool chain for avionics design, development, [74] R. Wilhelm et al., “Designing predictable multicore architectures for
integration and test,” in 1st Workshop on Avionics Systems and Software avionics and automotive systems,” in Workshop on Reconciling Perfor-
Engineering, 2019. mance with Predictability, Oct 2009.
[48] A. Davis, Software requirements: objects, functions, and states. [75] U. Dersch, “PLUS avionics whitepaper,” plc-tec, Whitepaper, 09 2018.
Prentice-Hall, Inc. Upper Saddle River, NJ, USA, 1993. [76] Real-Time Innovations (RTI), “Rti releases first do-178c certification
[49] W. Schamai et al., “Virtual verification of system designs against data package for its safety-critical connectivity platform,” 2015.
system requirements,” in Models in Software Engineering, J. Dingel and [77] S. H. VanderLeest, “Designing a future airborne capability environment
A. Solberg, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2011, (face) hypervisor for safety and security,” in 2017 IEEE/AIAA 36th
pp. 75–89. Digital Avionics Systems Conference, Sept 2017.
[50] A. Pop, D. Akhvlediani, and P. Fritzson, “Towards unified system [78] S. Toumassian et al., “Performance measurements for hypervisors on
modeling with the ModelicaML UML profile,” Simulation News Europe, embedded arm processors,” in Intl. Conference on Advances in Com-
vol. 17, no. 2, pp. 9–15, Sep. 2007. puting, Communications and Informatics, Sept. 21-24 2016.
[51] M. Tiller, Introduction to Physical Modeling with Modelica, ser. The [79] M. Narang et al., “Uav-assisted edge infrastructure for challenged net-
Springer Intl. Series in Engineering and Computer Science. Kluwer works,” in IEEE Conference on Computer Communications Workshop,
Academic Publishers, 2001. May 2017, pp. 60–65.
[52] M. Broy and K. Stølen, Specification and Development of Interactive [80] C. Watkins, C. Nilson, S. Taylor, K. Medin, I. Kuljanin, and H. Nguyen,
Systems. Springer, 2001. “Development of touchscreen displays for the Gulfstream G500 and
[53] I. H. Krueger, W. Prenninger, and R. Sandner, “Broadcast mscs,” Formal G600 Symmetry(TM) flight deck,” in 2018 IEEE/AIAA 37th Digital
Aspects of Computing, vol. 16, no. 3, pp. 194–209, 2004. Avionics Systems Conference, London, UK, Sept 2018.
[54] I. H. Krueger, “Distributed system design with message sequence [81] L. Ryon and G. Rice, “A safety-focused security risk assessment of
charts,” Ph.D. dissertation, 2000. commercial aircraft avionics,” in 2018 IEEE/AIAA 37th Digital Avionics
[55] F. Hölzl and M. Feilkas, “Autofocus 3: A scientific tool prototype Systems Conference, Sep. 2018, pp. 1–8.
for model-based development of component-based, reactive, distributed [82] H. Teso, “Aircraft hacking – practical aero series,” Apr. 2013, hITBSEC
systems,” in Proc. of the 2007 Intl. Dagstuhl Conference on Model- Conference 2013 Amsterdam.
based Engineering of Embedded Real-time Systems, ser. MBEERTS’07. [83] P. Berthier, C. Bresteau, and J. Fernandez, “On the security of aircraft
Berlin, Heidelberg: Springer-Verlag, 2010, pp. 317–322. communication networks,” in Critical Information Infrastructures Secu-
rity, G. D’Agostino and A. Scala, Eds. Cham: Springer Intl. Publishing,
[56] S. Kriebel, D. Raco, B. Rumpe, and S. Stueber, “Model-based engineer-
2018, pp. 266–269.
ing for avionics: Will specification and formal verification e.g. based on
[84] J. Juerjens, Secure Systems Development with UML, 1st ed. Springer,
broy’s streams become feasible?” in 1st Workshop on Avionics Systems
2005.
and Software Engineering, 2019.
[85] S. Crane, A. Homescu, S. Brunthaler, P. Larsen, and M. Franz, “Thwart-
[57] K. Hoelldobler and B. Rumpe, MontiCore 5 Language Workbench
ing cache side-channel attacks through dynamic software diversity,”
Edition 2017. Shaker Verlag, 2017, vol. 32.
in 22nd Annual Network and Distributed System Security Symposium,
[58] T. Nipkow, L. C. Paulson, and M. Wenzel, Isabelle/HOL — A Proof 2015, San Diego, California, USA, February 8-11, 2015, 2015.
Assistant for Higher-Order Logic, ser. LNCS. Springer, 2002, vol. [86] A. Burgess, The Executive Guide to Artificial Intelligence, 1st ed.
2283. Palgrave Macmillan, 2018.
[59] P. M. Duvall, S. Matyas, and A. Glover, Continuous Integration, ser. [87] I. Aizenberg, N. N. Aizenberg, and J. P. Vandewalle, Multi-Valued
Addison-Wesley Signature Books. Addison-Wesley, 2007. and Universal Binary Neurons: Theory, Learning and Applications.
[60] J. Humble and D. Farley, Continuous Delivery, ser. Addison-Wesley Springer Science and Business Media, 2013.
Signature Series. Addison-Wesley, 2010. [88] D. Silver et al., “Mastering the game of go without human knowledge,”
[61] M. Daum, J. Dörrenbächer, and B. Wolff, “Proving fairness and imple- Nature, no. 550, pp. 354–359, 2017.
mentation correctness of a microkernel scheduler,” Journal of Automated [89] C.-H. Cheng et al., “Neural networks for safety-critical applications —
Reasoning, vol. 42, no. 2, pp. 349–388, Apr 2009. challenges, experiments and perspectives,” in 2018 Design, Automation
[62] O. Kunčar, “Correctness of isabelle’s cyclicity checker: Implementability Test in Europe Conference Exhibition, March 2018, pp. 1005–1006.
of overloading in proof assistants,” in Proceedings of the 2015 Confer- [90] C.-H. Cheng, G. Nuehrenberg, and H. Ruess, “Maximum resilience of
ence on Certified Programs and Proofs, ser. CPP ’15. New York, NY, artificial neural networks,” in Automated Technology for Verification and
USA: ACM, 2015, pp. 85–94. Analysis, 15rd Intl. Symposium, D. D’Souza and K. N. Kumar, Eds.
[63] C. A. R. Hoare, “An axiomatic basis for computer programming,” Pune, India: Springer Intl. Publishing, 2017, pp. 251–268.
Communications of the ACM, vol. 12, pp. 576–580, 1969. [91] C.-H. Cheng, G. Nuehrenberg, H. Ruess, and H. Yasuoka, “Towards
[64] F. Kirchner et al., “Frama-c, a software analysis perspective,” In Formal dependability metrics for neural networks,” in Proc. of the 16th ACM-
Aspects of Computing, vol. 27, no. 3, 2015. IEEE Intl. Conference on Formal Methods and Models for System
[65] J. W. McCormick and P. C. Chapin, Building High Integrity Applications Design (MEMOCODE’18), 2018.
with SPARK. Cambridge University Press, 2015. [92] A. Kurakin, I. J. Goodfellow, and S. Bengio, “Adversarial examples in
[66] Y. Moy, E. Ledinot, H. Delseny, V. Wiels, and B. Monate, “Testing the physical world,” in Intl. Conference on Learning Representations
or formal verification: Do-178c alternatives and industrial experience,” Workshop Track, 2017.
IEEE Software, vol. 30, no. 3, pp. 50–57, May 2013. [93] K.-P. L. Vu, J. Lachter, V. Battiste, and T. Z. Strybel, “Single pilot
[67] C. Baier and J.-P. Katoen, Principles of Model Checking. The MIT operations in domestic commercial aviation,” Human Factors, vol. 60,
Press, 2008. no. 6, pp. 755–762, 2018, pMID: 30063410.
[68] J. Esparza and K. Heljanko, Unfoldings, ser. Monographs in Theoretical [94] S. M. Neis, U. Klingauf, and J. Schiefele, “Classification and review
Computer Science. An EATCS Series. Springer, 2010. of conceptual frameworks for commercial single pilot operations,” in
[69] K. L. McMillan, “Using unfoldings to avoid the state explosion problem 2018 IEEE/AIAA 37th Digital Avionics Systems Conference (DASC),
in the verification of asynchronous circuits,” ser. Lecture Notes in Sep. 2018, pp. 1–8.
Computer Science, G. von Bochmann David K. Probst, Ed., vol. 663. [95] Y. Lim, V. Bassien-Capsa, S. Ramasamy, J. Liu, and R. Sabatini, “Com-
Springer, 1992, pp. 164–177. mercial airline single-pilot operations: System design and pathways
[70] ——, Symbolic Model Checking. Kluwer Academic Publishers, 1993. to certification,” IEEE Aerospace and Electronic Systems Magazine,
[71] ——, “A technique of state space search based on unfolding,” Formal vol. 32, no. 7, pp. 4–21, July 2017.
Methods in System Design, vol. 6, no. 1, pp. 45–65, 1995. [96] G. A. Boy, “Requirements for single pilot operations in commercial
[72] C. Schroeter and V. Khomenko, “Parallel LTL-X model checking of aviation: A first high-level cognitive function analysis,” in Proc. of the
highlevel petri nets based on unfoldings,” ser. Lecture Notes in Computer Poster Workshop at the 2014 Complex Systems Design, 2014.
Science, R. Alur and D. Peled, Eds., vol. 3114. Springer, 2004, pp.
109–121.
[73] S. H. VanderLeest et al., “A framework for analyzing shared resource
interference in a multicore system,” in 2018 IEEE/AIAA 37th Digital
Avionics Systems Conference, Sept 2018.

Authorized licensed use limited to: University of Waterloo. Downloaded on May 30,2020 at 00:24:25 UTC from IEEE Xplore. Restrictions apply.

You might also like