Revision of Australian Defence Standard Def (Aust) 5679: Tony Cant, Brendan Mahony, Brenton Atchison

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Revision of Australian Defence Standard Def (Aust) 5679

Tony Cant∗ , Brendan Mahony∗ , Brenton Atchison†



Information Networks Division
Defence Science and Technology Organisation
PO Box 5111
Edinburgh 5108, South Australia
Email: [Tony.Cant, Brendan.Mahony]@dsto.defence.gov.au


Invensys Rail Systems Australia
10 Brandl Street
Eight Mile Plains 4113, Brisbane, Australia
Email: Brenton.Atchison@invensys.com

Abstract the further development of Def (Aust) 5679, with sup-


port from Invensys Rail Systems Australia (IRSA)
The Australian Defence Standard Def (Aust) 5679, under a contract established in June 2004. The Work-
entitled “The Procurement of Computer-Based ing Group consists of representatives from the DMO,
Safety-Critical Systems”, was originally published in DSTO and Invensys, as well as representatives from
August, 1998. Def (Aust) 5679 is currently undergo- Australian Army and Navy Technical Regulatory Au-
ing revision. This paper describes the standard and thorities. The revision process took the DefSafe re-
discusses the issues that have been highlighted during ports as a starting point, as well as the experience
the revision process. gained by the use of Def (Aust) 5679 on the Safety
Case for the autopilot for the Nulka decoy (Nulka is a
Keywords: Safety Critical Systems, Software, Assur- joint Australian/US ship defence system). The pro-
ance, Trust, Safety. cess involved close interaction between Invensys and
DSTO. The Working Group had regular meetings to
1 Introduction review recommendations and to discuss key procure-
ment issues. It should be noted that consensus on
The Australian Defence Standard Def (Aust) 5679, certain issues could not be reached by the Standard-
entitled “The Procurement of Computer-Based isation Working Group, and that some key recom-
Safety-Critical Systems”, describes procedures and mendations made by IRSA were not adopted in the
practices to be carried out during system development revised standard.
that are intended to provide sufficient assurance of This paper describes the structure of the revised
system safety (AEA 1998). It was written by the De- standard, as well as the revision process and issues
fence Science and Technology Organisation (DSTO) raised.
in response to what were (and still are) perceived
as deficiencies in extant military and civilian safety 2 Background
standards, following extensive consultation with gov-
ernment, industry and academic experts, both within The current revision of Def (Aust) 5679 is being car-
Australia and overseas. Def (Aust) 5679 was ori- ried out against the background of a major paradigm
ginally published by the former Army Engineering shift in military safety standards: the move to goal-
Agency (AEA) in August, 1998. based standards. These are standards that focus on
Def (Aust) 5679 has been applied to a number of broad principles and requirements to be placed on de-
Defence projects within Australia. This practical ex- velopers, and are not prescriptive about the processes,
perience has been very beneficial, and has exposed methods and tools that should be used during system
areas where there may be difficulty applying the development. This trend began in the USA under
standard. In 2001, the Defence Materiel Organisation the umbrella of acquisition reform, which was motiv-
(DMO) contracted the Software Verification Research ated by the desire to place responsibility for systems
Centre (SVRC) of the University of Queensland to (including safety) with defence contractors. Acquis-
provide support and advice on safety management to ition reform also involved the replacement of milit-
DMO projects, to review international safety stand- ary standards by civilian ones where possible. Thus,
ards, and to carry out a critical review of Def (Aust) for example, the US Military Standard MIL-STD
5679. This project (called “DefSafe”) resulted in sev- 882C (US DoD 1993) evolved into MIL-STD 882D
eral technical studies on Def (Aust) 5679, comparing (US DoD 2000) that focused on broad requirements
it to international safety standards, and highlighting for system safety and did not mandate a particular
areas where the standard needed improvement. process for achieving these requirements (although
The Standardisation Branch of the DMO has some guidance was provided in a non-normative an-
formed a Standardisation Working Group to guide nex to the standard). This standard is currently be-
Copyright 2005,
c Australian Government Department of De- ing revised; draft versions of MIL-STD-882E have
fence. This paper appeared at the 10th Australian Workshop apppeared (US DoD 2004).
on Safety-Related Programmable Systems, Sydney. Confer- More recently, the UK MoD has published Is-
ences in Research and Practice in Information Technology, Vol. sue 3 of the Interim Defence Standard 00-56 (UK
XX. Tony Cant, Ed. Reproduction for academic, not-for-profit MoD 2004) This is a goal-based standard that starts
purposes is permitted provided this text is included. from a number of high-level principles, and amplifies
Customer

System
Developer Safety Evaluator
Case

End Users

Auditor Certifier

Figure 1: Safety Management Agents

these into requirements on system developers. It is 3 Safety Management


very much less prescriptive than the previous issue of
the standard, and was produced following wide con- Def (Aust) 5679 provides requirements and guidance
sultation. One major change is that Safety Integrity on safety program management, including planning,
Levels (SILs), although used in earlier versions, are the safety lifecycle, monitoring and reviews.
no longer mandated by the standard. However, some A key feature of the safety management require-
guidance on the use of safety integrity requirements - ments is the identification of agents and their roles
to define the variation of the degree of rigour of sys- and responsibilities (see Figure 1). The Customer is
tem analysis with potential safety risk - is provided the organisation (usually a Systems Program Office)
in Annex C. within the DoD that is procuring the system. The
Def (Aust) 5679 is also based on fundamental Developer is the organisation (usually a commercial
safety principles. However, in the authors’ opinion one) that is responsible for producing the system re-
the time is not yet ripe for a purely goal-based safety quired by the Customer according to the Standard. In
standard, for the following reasons. Safety critical most cases, a Certifier will need to accept the system
systems engineering is still in its infancy and thus as being safe and suitable for service. The Auditor
in many cases system developers will need to seek provides oversight (on behalf of the Certifier) of the
further guidance beyond that given in a goal-based safety management process. The Evaluator is an in-
standard in order to produce a safety case. Such dependent organization responsible for checking the
guidance may be difficult to obtain and may also be detail and validity of the Developer’s arguments that
inconsistent with the spirit of the standard. Also, it the system will meet its safety critical requirements.
will take some time to determine whether or not the The Customer will often seek to involve End Users,
adoption of goal-based standards is successful. Thus who provide input on the safety aspects of user re-
it is not intended that Def (Aust) 5679 follow overseas quirements and on user interface and performance is-
trends and become purely goal-based. sues.
Rather, the objectives of the revision are to ad-
dress specific technical issues and feedback from ex- 4 Technical Safety Case
perience to assist its application to Australian De-
fence projects. The challenge is to make the stand- Def (Aust) 5679 is a system-level standard that
ard generic (so that it applies to as many systems as provides detailed requirements and guidance on the
possible), practical (with a clear description of the re- structure of the Safety Case for a safety-critical sys-
quirements on developers and evaluators) and effect- tem. It focuses on assurance activities: these are sys-
ive (in that application of the standard will provide tem development and analysis activities that provide
sufficient assurance). evidence that the system meets its safety require-
In the following sections, the structure of Def ments. The main mechanism for structuring the
(Aust) 5679 will be described in detail. Safety Case is to require a component-level design
(components may be implemented using software,
hardware or human operators).
Def (Aust) 5679 was one of the first international hence the conditional probability of an accident oc-
standards to mandate a technical safety case deliver- curring given that a System Hazard has been realised.
ing a structured argument and supporting evidence. As a guide, in the case where these conditional acci-
The technical safety requirements of Def (Aust) 5679 dent probabilities can be calculated, it is reasonable
are based on defining the necessary contents of the to assign strengths as follows:
safety case.
Figure 2 gives an overview of the Safety Man- • Low for probabilities greater than 10−2 ;
agement Process, showing the relationship between
the various stages of System Development and Safety • Medium for probabilities between 10−2 and 10−4 ;
Case Development. and
The key stages of Safety Case Development are
described in the following sections. • High for probabilities less than 10−4 .
It is very important to note that the standard does
5 Preliminary Hazard Analysis not allow probabilities to be assigned to the chance
of the System Hazard itself being realised (this is dis-
The aim of Preliminary Hazard Analysis (PHA) is to cussed further in Section 12.4).
define the system boundary and identify all possible Each System Hazard generates a System Safety
accidents, and their severities, that may be caused by Requirement (SSR), which expresses the fact that the
a combination of the states of the system, environ- hazard should not occur. To each SSR we assign one
mental conditions and external events. of six Levels of Trust (LOTs), named T1 ,... , T6 .
First of all, the system is described in terms of the The LOT is a function of the severity of the result-
interfaces between the system and its environment. ing accidents and the strength of external mitigating
The collection of System Functions defines the scope factors contributing to an accident sequence, and is
of the system (and hence what can be changed or de- calculated in accordance with Table 2. For each ac-
signed by the Developer), which will generally include cident sequence involving the corresponding System
operator procedures. In other words, the System Hazard, a default LOT is assigned to an SSR based on
Functions taken together define the System Bound- the severity of accident. If no external factors can be
ary. It is crucial that the System Boundary is clearly assigned to the accident sequences involving the cor-
understood and accepted by the Safety Management responding System Hazard, then the LOT remains at
Agents. Factors that are beyond the scope of sys- the default value for that severity. On the other hand,
tem control are called external and are determined if for each relevant accident sequence a strength of ex-
through reference to the Operational Context. ternal factors can be assigned, then the LOT can be
The next step is to identify all accidents that could reduced from its default value.
conceivably arise from a combination of system and The LOT is therefore a measure of how badly we
external conditions, as well as their severities, defined want the SSR to be satisfied; i.e. it is a measure of
as show in Table 1. the assurance required that the SSR is met by the
system.
SEVERITY DEFINITION
Catastrophic Multiple Loss of Life
Fatal Loss of Life 6 System Hazard Analysis
Severe Severe Injury
System Hazard Analysis (SHA) derives Component
Minor Minor Injury Safety Requirements (CSRs) for each component of
the system.
Table 1: Def (Aust) 5679 accident severities. First of all, the Developer must produce a high-
level description of the system, called the Component-
System Hazards are top-level states of the system Level System Design (CLSD). The CLSD consists of a
from which an accident could conceivably result, pos- decomposition of the system in terms of Components
sibly involving a further chain of events external to that combine to carry out the system functions. The
the system (called an accident sequence) as depicted CLSD must define how each Component is intended
in Figure 3. In this diagram, events [E] are shown to be implemented, that is, whether a function of the
as square boxes; accidents [A] are terminal events. Component is carried out in software, hardware, or
States are shown as circles - these may be hazard (H) by operator procedures.
states or states external (X) to the system. Hazard System Hazard Analysis details all possible
states should be defined in the most convenient way, Component-Level Hazards that could result in Sys-
even if this means that a number of hazards must be tem Hazards. Fault tree analysis may be used to do
realised to enable a given accident or that a number this. Each such hazard corresponds to a Component
of accidents may be enabled by a given hazard. Safety Requirement (CSR).
For each accident sequence, the Operational Con- The CSRs serve to focus the System Safety As-
text should be referenced to determine the strength surance effort on those parts of the system where it
of external factors that would perform hazard control is most needed. Once Component Safety Require-
or damage limitation functions. Such external factors ments have been documented they can serve as top-
must be assessed as one of Low, Medium, or High. In level safety specifications for system Components.
general, engineering judgement is used to make these
assessments. Such arguments must be made explicit 6.1 Hardware
in the Safety Case, and need to be carefully checked
by the System Safety Evaluator. This standard identifies two broad classes of hardware
In a limited number of cases (i.e. where the acci- Components: simple and complex. The behaviour of
dent sequence includes random external events such simple hardware may be directly determined by ex-
as environmental factors, the behaviour and condition haustive testing, while for complex hardware it is ne-
of hardware etc), it may be possible to ascribe a prob- cessary to infer behaviour through design analysis.
ability to the occurrence of these external factors, and
System Development Safety Case Development

System Definition
Preliminary
Hazard Analysis

Review &
Preliminary Evaluation
Design
System Hazard
Analysis

Review &
Evaluation

Safety Criticality
Assessment
Design Review &
Development Evaluation

Design Assurance

Implementation Review &


Evaluation

Implementation
Assurance
Integration and
Installation Review &
Evaluation

System Integration
Assurance

Review &
Evaluation

Final Review and


Post-Development
Evaluation

Figure 2: Def (AUST) 5679 Safety Management Process.

Examples of simple hardware include: mechanical Non-custom hardware Components present a chal-
interlocks, door latches, canisters, bulkheads, etc. For lenge for safety assurance. Adequate assurance can-
simple hardware Components, reliable, thoroughly not be gained through testing alone and detailed
tested and robust equipment must be used, and ana- design information may not be available. Assurance
lysed with respect to safety. must be gained by the use of generic, widely proven
Complex hardware Components may be further components.
classified as custom and non-custom components.
Custom hardware Components are those that are 6.2 Software
designed specifically for the system under devel-
opment or are otherwise limited in use, such as: Software allows the reuse of widely proven hardware,
application-specific integrated circuits (ASICs), pro- usually in the form of microprocessors, in the imple-
grammable logic controllers (PLCs) and program- mentation of a very wide class of system functions.
mable gate arrays (PGAs). It should be recognised that implementation of
The design of custom hardware Components must system functions by software means that it is easy to
use a structured architecture approach in which ele- introduce unmanageable complexity and hinder pre-
ments with well-defined interfaces are combined, and dictability of Component behaviour. Software does
must be supported by tools. not have to obey any physical laws and its behaviour
Non-custom hardware Components are general is not continuous. Behaviour can change radically
purpose devices, such as: microprocessors, disk-drives with small changes in input values. Thus, although
and other computer peripherals and PDAs. the purpose of software is to customise the behaviour
of general purpose computing hardware, it is concerns
System

E1 H1

E3 X2 A1

E2 H2

H3

E4 X3 A2

X1

Figure 3: Example accident sequences.

DEFAULT EXTERNAL
LEVEL MITIGATION
SEVERITY Low Medium High
Catastrophic T6 T6 T5 T4
Fatal T5 T5 T4 T3
Severe T4 T4 T3 T2
Minor T3 T3 T2 T1

Table 2: Def (Aust) 5679 Levels of Trust.

about the design of the software, rather than the re- to overall safety. Thus, operators must be considered
liability of the hardware, that tend to dominate. to be part of the system.
Contracts for the production of software Compon- The design involved in operator Components con-
ents must specify conformance to a software devel- sists of defining standard and emergency procedures
opment standard. Sound software engineering and that those operators must follow in interacting with
structured programming must be practised both in the system and the training they must undergo before
software design and coding. using it.
Software must be as simple as possible in structure In addition to the core activities of design, atten-
and in execution. The aim is to minimise sources of tion should be paid to the human-computer interface
unpredictability of software in operation, and maxim- and the procedures and training required of personnel
ise the opportunity for independent verification of its that operate, install or maintain parts of the system,
correctness by the use of tools. Other coding practice with special focus on the safety-related functions of
constraints may have to be applied as a consequence the system.
of the hazard analysis and consideration of the hard- Human factors analysis, especially the potential
ware environment in which the software operates. for operator induced hazards, must be carried out as
The programming language chosen shall allow the part of safety analysis and design of operator proced-
assurance of safety in accordance with the relevant ures and operator interfaces.
Safety Assurance Levels for critical requirements of Operator tasks should be designed to maximise
Components containing software, as discussed in Sec- performance through consideration of a number of
tion 9. factors, including the physical and cognitive demands
of the task, the conditions under which the task is per-
6.3 Operators formed, the nature of information presented to make
decisions and the mechanisms by which actions are
The way in which human operators interact with taken. The limitations and strengths of human per-
other Components of the system is very important formance should be considered in allocating operator
tasks and designing their interaction with a computer
system. 8.1 Component Models
In order to reason about the behaviour of a Compon-
7 Safety Criticality Assessment ent and show that it meets its CSRs, a model of the
Component Design must be formulated. The level of
The next stage of the Safety Case is called Safety Crit- formality of the language used to formulate the model
icality Assessment. This assigns a Safety Assurance of a Component is specified in Table 3 (the SAL used
Level (SAL) to each CSR. The SAL of a CSR pre- is the highest for the CSRs associated with the com-
scribes the system development and safety analysis ponent).
techniques that need to be applied during develop-
ment to help ensure that a given system Component
will achieve the required level of assurance. As with 8.2 Safety Requirements Specification
LOTs, there are six SALs S1 , ... , S6 . The Component Safety Requirements identified by
The SAL that is assigned to a CSR may be reduced System Hazard Analysis shall be specified as asser-
by the application of protective measures (such as in- tions about the Component’s interface with other
terlocks) in the system design that reduce the chance Components. The level of formality of the specific-
of the corresponding hazard resulting in an accident. ation language used is prescribed in Table 3.
The use of protective measures means that there are It is sufficient to express Component Safety Re-
multiple, independent CSRs that prevent the occur- quirements with a SAL of S1 as informal English lan-
rence of an accident, either by prevention of a single guage statements. For SALs S3 or higher, Component
System Hazard or prevention of multiple System Haz- Safety Requirements are expressed in a formal spe-
ards that are combined within the same accident se- cification language. Safety requirements with a SAL
quence. of S2 require some form of semi-formal specification,
If no specific protective measures have been taken, which involves the use of structured English language
then the SAL of a given CSR is the same as the LOT statements, possibly augmented with diagrams and
of the SSR from which the CSR was derived. The mathematical notation.
default SAL of a CSR may be reduced if protective
measures are in place that prevent the occurrence of
a corresponding Component Hazard from resulting in 8.3 Design Verification
an accident. Design Verification proofs must be carried out for
Although protective measures are an important each system Component against its CSRs. The degree
method for containing the consequences of failures of formality of the proofs is prescribed in Table 3.
within the system, there is a limit to their use. A For the SALs S1 or S2 , it is sufficient that such
certain amount of trust may be placed in protect- proofs be expressed using informal clear and logical
ive measures and the corresponding arguments in the English arguments. For SALs S5 or S6 , proofs need to
Safety Case as to how the assurance of safety will be be expressed using a formal mathematical logic. Such
increased by these measures. However, the indiscrim- proofs are called formal. The SALs S3 and S4 require
inate use of arguments for safety that rely on protect- some form of rigorous proof. Such proofs are mathem-
ive measures can result in components being assigned atically convincing but may omit some of the detailed
an assurance level far below their corresponding con- justification. For example, the proof could consist of a
tribution to possible hazards. high-level formal argument but with lower-level steps
The amount of trust that is gained by the use of justified informally.
such measures is not offset by the trust that is lost
by excessive reduction of SALs. This is because the
assurance gained by activities carried out at higher 9 Component Implementation Assurance
SALs (such as design modelling, verification etc) out-
weighs the assurance gained by the use of protect- The aim of Component Implementation Assurance is
ive measures and hazard mitigation arguments. Ac- to provide detailed evidence that the implementation
cordingly, the SAL of a CSR shall be no less than of a given component will meet its CSRs. Assurance
two levels lower than the LOT Level of the SSR from of system safety is established through the applica-
which it was derived. tion of specific development and analysis activities
Def (Aust) 5679 provides detailed guidance on the for each System Component, according to the SALs
various kinds of protective measures, and how SALs of associated CSRs.
can be reduced by their use. Such measures include: The requirements for software components are
functional redundancy, data redundancy, design re- summarised in Table 4 and include the following:
dundancy, safety kernels, etc. 1. Defensive programming techniques, which in-
clude the use of safe subsets of high-level lan-
8 Component Design Assurance guages, in which only program constructs with
well-defined semantics are allowed.
The aim of Component Design Assurance is to analyse
Component Design and the process of Component 2. Flow analysis to study aspects of code such as
Development, and provide sufficient assurance that, control flow, information flow and data use ana-
if the Component is implemented as designed, the lysis. It must be supported by the use of static
Component Safety Requirements are met. analysis tools.
Assurance of system safety is established through 3. Code specification, in which the CSRs are decom-
the application of specific development and analysis posed into corresponding Unit Safety Require-
activities for each System Component according to ments (USRs) representing critical properties of
the SALs of associated CSRs (see Table 3). source code units. The specification of USRs may
be presented into informal, semi-formal or formal
language, as for the CSRs.
S1 S2 S3 S4 S5 ,S6
System Component Model Informal Informal Semi-Formal Formal Formal
Specification of CSRs Informal Semi-formal Formal Formal Formal
Design Verification Informal Informal Rigorous Rigorous Formal

Table 3: Def (Aust) 5679 Safety Assurance Design Attributes.

S1 S2 S3 S4 S5 S6
Defensive High Level High Level Safe Safe Safe Safe
Programming Language Language Subset Subset Subset Subset
Flow Analysis Optional Optional Yes Yes Yes Yes
Code Informal Informal Semi- Semi- Formal Formal
Specification Formal Formal
Code Informal Informal Rigorous Rigorous Formal Formal
Verification Object
Code
Software Basic Medium Medium High High High
Safety Testing

Table 4: Def (Aust) 5679 Safety Assurance Level Attributes for Software.

4. Code verification to prove that each software unit back into system and Safety Case development. It is
satisfies its USRs. The proofs may be presented important to avoid the extreme situations where, on
using informal, rigorous or formal arguments. the one hand, System Safety Evaluators become part
of the development process (and thus compromise in-
5. Software safety testing, broadly classified into dependence) and, on the other hand, Evaluation is
unit testing and integration testing. The pur- done only at the very end of development (this could
pose of unit testing is to demonstrate that soft- be very costly if a safety issue is not addressed in the
ware units satisfy their corresponding USRs. The Safety Case, and could have been picked up by the
purpose of integration testing is to show that the Evaluator).
complete component satisfies its corresponding The Auditor must take the lead in resolving any
CSRs. disputes that may arise between the Customer, De-
In practice, the assurance evidence may combine veloper and Evaluator.
to form a sound argument. For example, justification
that the software safety testing need cover only a spe-
cific part of the software, may be provided by a flow 12 Issues Addressed
analysis showing that the remaining part does not
interact, and defensive programming demonstrating The Def (Aust) 5679 revision process has identified
that the entire software is exception free and cannot a number of potential issues with the standard that
indirectly modify critical variables. have been debated at length. Many of these issues
represent fundamental challenges for computer-based
safety engineering. Others highlight certain key prin-
10 System Integration Assurance ciples of Def (Aust) 5679 that differentiate it from
some other safety standards. This section discusses
The aim of System Integration Assurance is to provide some of the issues and the current status of their res-
sufficient assurance that the System Safety Require- olution.
ments are met by the integrated and installed system.
System Safety Testing is performed on the in-
tegrated System to demonstrate that the collection 12.1 Non-Development Systems
of Components satisfy System Safety Requirements. The most pressing issue with the application of Def
System Safety Testing should exercise the system in (Aust) 5679 is the treatment of Non-Development
accordance with its expected operating demands. It Systems, which are systems that have been already
may be partially performed in a simulated environ- developed overseas for other Defence customers and
ment prior to Installation, or performed on the in- potentially modified to suit local conditions and re-
stalled system, when safe to do so. quirements. Procurement of such systems is necessary
Installation Safety Assurance provides evidence of practice for the Australian Department of Defence.
the safety of each System Installation, including the Non-Development Systems may have been developed
manufacture and assembly of all components, use of in accordance with an international safety standard or
any customisation data, and the installation itself. no safety standard at all, but rarely with Def (Aust)
5679 compliance in mind. Convincing evidence may
11 Evaluation be difficult to elicit, and such systems present great
challenges for procurement.
Def (Aust) 5679 requires that an independent System The procurement of non-development systems
Safety Evaluation be carried out to assess the tech- poses three major issues for Def (Aust) 5679. First,
nical validity of the Safety Case. Effective System the avoidance of probabilistic targets and ALARP
Safety Evaluation is absolutely essential for the assur- (see Section 12.3) means that a safety risk assessment
ance of safety; it relies on diversity and independence must be conducted in Def (Aust) 5679 terms in order
in the review of technical arguments. for the standard to be applied. Second, since the Def
The Evaluation is carried out in well-defined (Aust) 5679 assurance requirements are deliberately
stages, after each of the Safety Case activities de- more rigorous than other standards, it is likely that
scribed above. The results of evaluation then feed non-development systems will not meet it, even if they
have been accepted overseas. Third, Def (Aust) 5679 use their own risk tolerability criteria; they may de-
currently provides no means of incorporating field ser- cide not to use an independent evaluator, and so on.
vice evidence into a safety argument. The revision The ability to tailor standards is easily abused. The
process has suggested a means by which this could authors do not believe that tailoring is appropriate
be achieved and has applied it to the limited scope of and it is not permitted in Def (Aust) 5679.
non-custom hardware components. However, broader
application requires further consideration. 12.4 Probabilities
The Standardisation Working Group has acknow-
ledged that the procurement of Non-Development Def (Aust) 5679 makes only limited use of probab-
Systems presents fundamental questions of the role of ilities. They are permitted only when assessing the
an indigenous safety standard and its enforcement in chance of external events as contributions to accident
the Australian Defence environment. These questions sequences (and may then be used as a basis for re-
are equally applicable to the role of standards more ducing LOTs). However, the use of probabilities for
generally and the ability of a procurement authority system hazards in computer-based systems is prohib-
to enforce standards in the modern age of global sys- ited by the standard.
tems development. In part, this is a likely motivation There are several reasons for avoiding the use of
for the trend towards goal-based standards seen else- probabilities. First, it is considered to be inappro-
where. priate to associate probabilities to design failures, in-
cluding conditions arising from software components.
12.2 Strength of Assurance Second, it is believed that probabilities can be used to
obscure unsound safety arguments arising from erro-
The strength of the assurance demanded by safety neous or optimistic data and independence claims. A
standards ranges considerably: from the lowest as- third reason is that the use of failure probabilities may
surance, such as MIL-STD 882C (DoD93 1993), cause confusion between claims of system reliability
through intermediate assurance (such as IEC 61508 and evidence that the system is safe. A fourth reason
(IEC 2005) and RTCA DO-178B (RTCA 1992)), to is that the standard does not require full analysis of
very high assurance, such as the original version of subcomponents such as processors and operating sys-
the UK DefStan 00-55, which contained extremely tems, making it difficult to achieve accurate probabil-
demanding requirements for Safety-Critical Software. istic calculations taking account of such subsystems.
The strength of assurance that is adopted is to some Finally, it is not clear that the state-of-the-art per-
extent a question of engineering judgement about mits the development of probabilistic arguments to
what best practice should be. Def (Aust) 5679 adopts the level of formality required by other parts of the
a strength of assurance intended to be high enough standard.
to give confidence in system safety, but not so high as One limitation arising from the avoidance of prob-
to place unreasonable demands on developers. The abilities is that the standard provides no framework
use of high-assurance techniques can also provide in- for assessing events such as the failure of physical ma-
creased understanding of system requirements and terials and certain human behaviours, which are best
design; thus, it can lead to cost savings for later stages modelled in probabilistic terms.
of the project. It is accepted that the treatment of such random
The revision process raised the issue of whether the events is an important aspect of safety engineering.
Def (Aust) 5679 assurance requirements represent a However, in light of the concerns raised above, it
standard that can be met at acceptable cost, particu- is not yet clear how best to address these matters.
larly given the propensity for procurement of overseas The main options are either to associate probabil-
and previously developed system. However, in view istic targets with SSRs and flow the targets top-down
of the above, and on the advice of the Def (Aust) through all the system design stages, or else to syn-
5679 editor, the Standardisation Working Group has thesise probabilistic estimates bottom-up in a separ-
decided to make no major changes to the assurance ate stage following final implementation. The former
requirements. seems to place undue constraints on design choices
and also to risk confusion between assurance and re-
12.3 Risk Tolerability liability engineering activities. The latter seems to
offer too little guidance to the design process. The
Def (Aust) 5679 does not use risk tolerability criteria. view of the Def (Aust) 5679 editors is that the latter
In particular, it does not invoke the ALARP (As Low option is preferable as it is of great importance that
as is Reasonably Pr icable) principle so familiar in failure rates not be equated with assurance levels.
the UK context through the Health and Safety at
Work Act (HMSO74 1974). The ALARP principle 12.5 System Integration and Deployment
states that there is a region of risk between strictly ac-
ceptable and strictly unacceptable levels where risks The current revision of Def (Aust) 5679 provides
should be reduced where practicable. ALARP is in- scope for the analysis and assurance of systems up
timately bound up with issues such as how to measure to the development of system components. However,
risk, what weight to give risk reduction activities and experience from early application of the standard re-
ultimately becomes bound up with cost estimation. vealed the need for the technical safety case to ex-
It is also problematic for software. Def (Aust) 5679 plicitly address system integration and deployment.
makes no use of it and, as a consequence, does not Accordingly, the revision process has added require-
allow the acceptance of specific risk levels by a higher ments and guidance to ensure that the safety case
authority. include test evidence from system integration and ac-
A related issue is that Def (Aust) 5679 is not in- ceptance testing. For this purpose, the standard has
tended to be tailored. The processes described in it distinguished the development of an initial system
are mandatory. Many standards do allow for tailor- from the manufacture and deployment of multiple
ing: developers or procurement agencies may decide system instances, including the potential for custom-
to lower assurance requirements; they may decide to isation of each instance. While the generic system
safety case must demonstrate safety of the general
design, each instance need only demonstrate safety of 13.3 Tailoring
the manufacture, installation and customisation.
There were several clauses in the published version
that gave the safety management agents the scope
12.6 Operation and Support to substantively modify the strength of assurance
Def (Aust) 5679 deliberately targets the procurement delivered by application of Def (Aust) 5679. Such
of systems and does not cover the safety of systems clauses allowed scope for significant abuse of the spirit
once accepted into service. For this strategy to be ef- of the standard and have been removed.
fective, the standard must clearly identify those oper-
ating and support procedures necessary to maintain 13.4 Changes in Terminology
system safety in service. The system must be de-
signed to minimise the risk that these procedures will A number of changes of terminology have been made
be violated, and the standard must ensure that re- in the revised version of Def (Aust) 5679. Some of
sponsibility for meeting the safety procedures is trans- the more significant ones are listed in Table 5.
ferred to the operational body when the procurement
safety case is accepted. While these issues were ad- 14 The Next Step
dressed to some extent in the original Def (Aust)
5679, the revision process added requirements and The intention is to issue a public draft revision of
guidance to make this more explicit. However, a more Def (Aust) 5679 late in 2005 for review and comment
holistic view of Defence procurement has been devel- by Defence and industry, as well as the international
oping over recent years, and Def (Aust) 5679 will need safety community. In particular, comment will be
to pay greater attention to whole of lifecycle issues in sought from members of The Technical Cooperation
the future. Panel (TTCP) Focus Area JSA-TP-4-3 on Safety-
Critical Systems. This is a collaborative group in-
13 Key Revisions volving Defence Research and Development and pro-
curement representatives from the UK, USA, Aus-
In this section we discuss some of the key changes to tralia and Canada. Defence safety standards are reg-
the standard as a result of the revision process. ularly discussed in this forum.
These comments will then be collated and presen-
ted to the Standardisation Working Group for its fur-
13.1 Accident Sequences ther consideration. It is expected that the revised
The analysis of accident sequences, according to the standard will be published by the end of 2005.
published version of the standard, was a matter of For further information on Def (Aust) 5679, please
some concern in practice. contact Dr Tony Cant (DSTO).
A particular point of confusion was the question of
correspondence between accidents and hazards. For 15 Acknowledgements
example, is it legal to have more than one hazard lead-
ing to a given accident (via differing sequences) or to The DSTO contribution to this work was carried out
have more than one hazard leading to the same acci- under the Critical Systems Development Task JTW
dent? These issues have been clarified in the revised 04/061, which is sponsored by the Standardisation
version. Branch of the Defence Materiel Organisation. The
The published version of the standard did not ISRA contribution was carried out under contract to
allow qualitative estimates of external mitigating the DMO.
factors when determining the LOT required for a
given SSR. This was of concern in practice for two
reasons. Firstly, it denied the value of and failed to References
encourage the use of external mitigating factors that
cannot be clearly quantified. Secondly, it could be Army Engineering Agency (1998), Def (Aust) 5679:
seen as encouraging the attribution of probabilities The Procurement of Computer-Based Safety-
that could not be justified, a practice in contraven- Critical Systems.
tion of the general spirit of Def (Aust) 5679. US Department of Defense (1993), Military Stand-
ard MIL-STD-882C: System Safety Program Re-
13.2 Assurance Framework quirements.
The development and analysis processes required for Department of Defense (2000), Military Stand-
each safety assurance level were reviewed in detail and ard MIL-STD-882D: Department of Defense -
some small, but significant, improvement were made. Standard Practice for System Safety.
For example, the notions of semi-formal specification
and rigorous proof were a point of some confusion and US Department of Defense (2004), Draft Military
have been more carefully defined in the revision. The Standard MIL-STD-882E: Department of De-
role and practice of testing in software assurance has fense Standard Practice for System Safety.
been significantly revised, leading to a more stream- UK Ministry of Defence (2004 ), Safety Management
lined process that is better in line with modern testing Requirements for Defence Systems: Part 1 (Re-
practices. The contribution of physical reliability to quirements), Part 2 (Guidance).
system safety has been more clearly explained.
It should be noted, however, that the overall Her Majesty’s Stationery Office (1974), Health and
strength of assurance has not changed in the revision, Safety at Work etc Act.
as discussed above.
RTCA Inc (1974), Software Considerations in Air-
borne Systems and Equipment Certification.
Original Term New Term Reason
Safety Critical Safety Critical The term “computer-based” was dropped to
Computer-Based Systems emphasize the system nature of the standard,
Systems and to remove the focus on software.
Safety Integrity Safety Assurance The notion of Safety Integrity Level is used
Level Level in many standards and has a rather differ-
ent meaning from that invoked by Def (Aust)
5679, which uses the term as a measure of
strength of assurance.
Safety Integrity Safety Criticality Again, the term “safety integrity” was felt to
Assessment Assessment be misleading
Post-Development Non-Development The term “post-development” could be taken
Systems Systems as meaning ”after the development of the cur-
rent system”, which was not what was inten-
ded.

Table 5: Terminology.

International ElectroTechnical Commis-


sion (1974), Functional Safety of Elec-
trical/Electronic/Programmable Safety-Related
Systems - IEC 61508.

You might also like