Академический Документы
Профессиональный Документы
Культура Документы
Sommersemester 2001
%once&ts must #e clear in order to ena#le easy and uniform use in engineering 'f conce&ts are unclear( then
There are various conce&ts of risk and ha+ard ,hich are effective for engineering &ur&oses( and -hich are not.
23 !"r# 2005
They 0-ork1
They are a&&lica#le to the domain of interest They ena#le &assa#le assessments of risk and safety"critical failure They can #e used -ithin an engineering organisation 2ood arguments e3ist concerning a&&lica#ility and correctness
Summary4 good engineering means kno-ing -hat methods are a&&lica#le( -here( -hy and ho-.
$ 3
23 !"r# 2005
system( environment( #oundary( o#5ects( fluents( state( state change( event( #ehavior( near and far #ehaviors( necessary causal factor plus Likelihood( Severity Likelihood( %onse8uences
6ccident
7a+ard
9isk
$ %
23 !"r# 2005
$ 23 !"r# 2005 5
+e !oivre
6#raham de :oivre(
in the Philoso&hical Transactions of the 9oyal Society 0The 9isk of losing any sum is the reverse of /3&ectation< and the true measure of it is( the &roduct of the Sum adventured multi&lied #y the Pro#a#ility of the Loss.1 Severity4 0the Sum adventured1
$ 23 !"r# 2005 6
+e !oivre$ -is.
$ 23 !"r# 2005 ,
Accident4 0an undesired and un&lanned )#ut not necessarily une3&ected* event that results in )at least* a s&ecified level of loss.1 Hazard4 0a state or set of conditions of a system ... that( together -ith other conditions in the environment of the system ...( -ill lead inevita#ly to an accident1
defined -ith res&ect to the environment -hat constitutes a ha+ard de&ends on the #oundary of the system
$
23 !"r# 2005
Hazard )cont>d*
Severity )or damage*4 0the -orst &ossi#le accident that could result from the ha+ard given the environment in its most unfavora#le state1 Likelihood of occurence Hazard level4 0com#ination of severity and likelihood of occurrence1
Risk: 0the ha+ard level com#ined -ith )1* the likelihood of the
ha+ard leading to an accident ... and )2* ha+ard e3&osure or duration1
23 !"r# 2005
6ccident
an un-anted event a system state( -hich in com#ination -ith the most unfortunate environment state( results inevita#ly in )is a sufficient causal factor of* an accident Level of loss )on a ratio scale*
7a+ard
Severity
9isk
= &)7*.&)%3 ? 7*.S)%3*
%34 an accident that results from@through 7 S)%3*4 the severity of the$ accident %3
10
23 !"r# 2005
6 commercial aircraft encounters thunderstorm tur#ulence -hich causes loss of control and #reaku&
,hen the environment contains such tur#ulence( and the aircraft is flying( then an accident is inevita#le 't follo-s that flying states of the aircraft are ha+ard states
/nvironment state
'n this e3am&le( as in the game of golf or of real tennis( the 0ha+ard1 is more intuitively an environmental state
2lo#al State )Aackson 1BBC< Sim&son D Stoker 2002* '/% E1C0F4 0&otential source of harm1
seems to allo- systemGenvironment state )global state* #ut then( it seems to allo- lots of things
$ 11
23 !"r# 2005
7o- do = &)7*.&)%3 ? 7*.S)%3* and = &)%3*.S)%3* com&are. 71( 71( ... (7k a collection of mutually e3clusive ha+ards such that each accident ha&&ens through one of them Then #y a #asic calculation in conditional &ro#a#ility &)%3* H = &)7i*.&)%3 ? 7i* Thus &)%3*.S)%3* H = &)7i*.&)%3 ? 7i*.S)%3* 6nd summing over all %3 yields the result
)9e&eat*4 71( 71( ... (7k a collection of mutually exclusive hazards such that each accident happens through one of them ,ithout this assum&tion( the sums may not #e the same $
12
23 !"r# 2005
7arm
0&hysical in5ury or damage to the health of &eo&le either directly( or indirectly as a result of damage to &ro&erty or to the environment1 0&otential source of harm1 0ha+ardous situation -hich results in harm1 0circumstance in -hich a &erson is e3&osed to ha+ard)s*1
$ 13
7a+ard
7a+ardous event
7a+ardous situation
23 !"r# 2005
9isk
0com#ination of the &ro#a#ility of occurrence of harm and the severity of that harm1 0risk -hich is acce&ta#le in a given conte3t #ased on the current values of society1 0freedom from unacce&ta#le risk1
Tolera#le 9isk
Safety
$ 23 !"r# 2005 1%
0ha+ardous event1 comes close( #ut is a 0situation1 #ut US aviation regs )1I %F9 FJ0 KFJ0.2* allo- an accident to #e significant aircraft damage alone similarly -ith US6F %lass 6 misha&s )the severest sort*
23 !"r# 2005
9isk
'f severity is 8uantitative( does@can 0com#ine1 mean 0multi&ly1. 'f so( then risk is defined here to #e a multi&lication 'n de :oivre D Leveson( it is a sum
$ 23 !"r# 2005 16
't is certain ' shall suffer some degree of harm -hile using my #icycle )from a trivial scratch from a &art once a month( to falling off once a decade( to #eing run over* The &ro#a#ility of harm is 1 Severity is varia#le from trivial to catastro&hic ,hich 0severity1 do ' use. %all it S 7o- do ' 0com#ine1 S -ith 1. 't cannot mean the actual harm that -ill in fact occur( since that -ould render the conce&t unusa#le for calculation in advance( as '/% E1C0F re8uires during system develo&ment )0/U% risk1( 0tolera#le risk1( 0residual risk1*
$ 1,
23 !"r# 2005
2ood definitions )good &rograms* define terms )varia#les* #efore they use them ) ef"use test( used a lot in static analysis of &rograms* Usa#le definitions try to #e &recise reduce or eliminate am#iguity limit the num#er of undefined conce&ts #e clear to the intended inter&reters :y o&inion4 '/% E1C0F does not do -ell on these criteria( similar to many )#ut #y no means all* engineering standards
$ 18
23 !"r# 2005
System Lifecycle Fuctional Safety 9isk and 9isk 9eduction System Su#division Safety 'ntegrity Level )S'L* 6s Lo- 6s 9easona#ly Practica#le )6L69P*
$ 23 !"r# 2005 11
$ 23 !"r# 2005 20
Overall plannin8 OveralI 6 operation , and (aintenance plannin8 Overall safety validation plannin8
1
OveralI
10
11
-ealisation
)see /@/@P/S safety lifecycle*
-ealisation
-ealisation
12 13
Back to appropriate overall safety lifecycle phase (odification 15 Overall and retrofit
Overall operation)
$ 23 !"r# 2005
16
+eco((issionin8 or disposal
21
-ealisation
12
13
1 % E9E9:E' inte8ration
15
23 !"r# 2005
22
13
1%
:E inte8ration /hardware9software0
16
$ 23 !"r# 2005 23
The *ifecycle
Lne needs a lifecycle model The E1C0F lifecycle model is as good as any and more detailed than most 7o-ever( there is no guidance on ho- to fit it to a ty&ical system develo&ment lifecycle
$ 23 !"r# 2005 2%
9e8uirements
%ode develo&ment( Testing 'ntegration( 'ntegration"Testing Further develo&ment according to ne- re8uirements( :odification through error correction and failure correction
'm&lementation
0:aintenance1
ecommissioning
$ 25
23 !"r# 2005
,e need to harmonise the '/% E1C0F lifecycle model and the ty&ical system develo&ment lifecycle model used in a firm &resumed to #e straightfor-ard( #ut ho- do -e kno-. ,ho has done it. There are three sorts of different re8uirements in '/% E1C0F )Fenton@Meil( 1BBF* For the final &roduct )the S% system* For documentation
S&ecifications at the various levels 6nalysis and re&orting documents( e.g. the Safety %ase checks and sign"offs to #e conducted #y 8ualified &ersonnel
$ 26
For resources
23 !"r# 2005
Functional Safety
Safety &ro&hyla3is restricts itself to safety functions Safety functions are actions( that are 0intended to achieve or maintain a safe state for the /U%( in res&ect of a s&ecific ha+ardous event1 9ecall that a ha+ardous event results in harm. 'f harm is to #e avoided #y means of the safety function( then the function should inhi#it the s&ecific ha+ardous events -hich are &recursors of the harm
23 !"r# 2005
9isk 9eduction
There is no such thing as 0Oero 9isk1 The Safety Functions )SF* are concerned -ith risk reduction
There is an E ! risk4 0risk arising from the /U% or its interaction -ith the /U% control system P/U%%SQ1 There is a tolerable risk There is a residual risk4 0risk remaining after &rotective measures have #een taken1 evelo&ers must assess the /U% risk and the tolera#le risk )to calculate the re8uired safety integrity level( S'L* as -ell as the residual risk( -hich must #e as lo" as reasonably practicable )6L69P*
$ 28
23 !"r# 2005
/8ui&ment under control )/U%* /U% control system )/U%%S* Safety"9elated System )S9S* The /U%%S can #e classified as an S9S or not )#ut the criterion( in clause ;.C.2.I( is a logical tautologyNN*
Safety"9elated System
im&lements the re8uired safety functions ....... and is intended to achieve Pin &ossi#le com#ination -ith othersQ the necessary safety integrity for $the re8uired safety functions1
21
23 !"r# 2005
-is. -ed6ction
-esid6al ris. Tolera;le ris. E@C ris.
Increasin8 ris.
-is. red6ction achieved ;y all safety<related syste(s and e=ternal ris. red6ction facilities $
23 !"r# 2005 30
9isk of /U%@/U%%S -ithout S9Ss 9isk of /U%@/U%%S G S9Ss 6cce&ta#le 9isk )socially derived*
$ 23 !"r# 2005 31
Concepts 5$ 'I*
/ach S9S is assigned a S'L( -hich re&resents the &ro#a#ility that the S9S fulfils its safety function)s* That is( the S'L of an S9S re&resents o#5ectively the relia#ility of its safety function)s* )a product re#uirement* The S'L is assigned according to the re8uired risk reduction )from /U% risk at least to the tolera#le risk* 6 8uantitative difference is made #et-een
evelo&ment of an S9S -ith a designated S'L re8uires a certain develo&ment &rocess )a process re#uirement*
$ 32
23 !"r# 2005
'I*s) contin6ed
S'L )cont>d*
' shall ignore the difference #et-een lo-"demand and high" demand modes Four levels of increasing relia#ility )S'L 1 R S'L I*
'm&licitly five( -ith S'L 0( a#out -hich nothing is said /ach level re8uires a relia#ility of 10S)")nG1** to 10S)"n* dangerous failures &er hour@&er demand 7ighest recognised level ist nHF )S'L I( continuous mode*
$ 23 !"r# 2005 33
High demand or continuous mode of operation (Probability of a dangerous failure per hour) 10-9 to < 10-8 10-8 to < 10-7 10-7 to < 10-6 10-6 to < 10-5
$ 23 !"r# 2005 3%
The distinction #et-een lo-"demand and high" demand modes may -ell disa&&ear in the ne3t release of E1C0F )Simon Bro-n( 200C* 6 S'L is valid for a particular S$%component in a particular system&environment given a )socially" determined* particular tolerable risk
7o-ever( organisations such as the TTUs are starting to 0certify1 com&onents inde&endent of s&ecific a&&lication There is a real danger that a S'L -ill #e seen as a &ro&erty of the com&onent( -hich it is not )9edmill 2000( 7amilton"9ees( 1BBB*
$ 35
23 !"r# 2005
S'L 1 target failure rates are already #eyond &ractical verification )Little-ood"Strigini 1BBJ( Butler"Finelli 1BBJ* S'Ls 1"I su#divide a &ro#lem s&ace in -hich there is no sensi#le distinction to #e made amongst a&&lica#le develo&ment and assurance methods For many recommended methods( there is little or no evidence that they reduce failure rates There is increasing evidence that those methods -hich do reduce failure rates also save money4 they should #e used at any S'L
$ 36
23 !"r# 2005
so the focus shifts from &roviding ade8uate safety )&roduct* to fulfilling the recommendations of the standard )&rocess* But there is little correlation #et-een &rocess &ro&erties and safety
Focus shift from &roduct to &rocess does not hel& safety )Mote4 There are conce&ts of S'L in other standards -hich suffer from only some of these &ro#lems. PBL*
$ 3,
23 !"r# 2005
Less than one dangerous failure every 10SF o&"hours )But more than one dangerous failure every 10SB o&" hoursNN aft.*
%ommercial aviation )-hich re8uires lo-er failure rates for certain critical su#systems( and the general history suggests this can #e achieved* The automo#ile industry )-hich has a real re8uirement of S9S relia#ility of u& to 10S10 o&"hours &er failureNN*
$ 38
23 !"r# 2005
Concepts 6$ A*A-:
To calculate the re8uired risk reduction( one must use the 6s Lo- 6s 9easona#ly Practica#le )6L69P* &rinci&le Lrigins4 /nglish la
Lord 6s8uith( 1BIB significantly reinforced4 Lord %ullen )1BFB*( Pi&er 6l&ha oil &latform fire investigation 6cce&ta#le4 so lo- that it can for all &ractical &ur&oses #e ignored 'ntolera#le4 so high as to #e unacce&ta#le in all circumstances The 6L69P region4 the region #et-een acce&ta#le and intolera#le( in -hich the system develo&er is re8uired to reduce the risk to #e 0as lo- as reasona#ly &ractica#le1 $
31
23 !"r# 2005
A*A-:
6L69P )cont>d*
'n legal cases( the UV 7S/ regards the 6L69P &rinci&le as having #een fulfilled if a develo&er is a#le to esta#lish that a system -as develo&ed in accordance -ith '/% E1C0F ):ark Bo-ell( UV 7S/( mailing"list comment( 200I* So it seems as if '/% E1C0F re8uires 6L69P( #ut to conform -ith 6L69P one needs only to do everything else Logically( this makes 6L69P redundantNN
23 !"r# 2005
%0
6s the risk is reduced( the less( &ro&ortionately( it is necessary to s&end to reduce it further to satisfy 6L69P. The conce&t of diminishing &ro&ortion is sho-n #y the triangle.
Ae8li8i;le ris.
$ 23 !"r# 2005 %1
Frequency Frequent Probable Occasional Remote Improbable Incredible Catastrophic I I I II III IV Critical I I II III III IV
$ 23 !"r# 2005 %2
Interpretation Intolerable risk Undesirable risk, and tolerable only if risk reduction is impracticable or if the costs are grossly disproportionate to the improvement gained Tolerable risk if the cost of risk reduction would exceed the improvement gained Negligible risk
$ 23 !"r# 2005 %3
9isk %lasses ' and 'U fit -ith 6L69P 9isk %lasses '' and ''' don>t o#viously fit -ith 6L69P
'n the region in -hich 9isk %lasses '' and ''' a&&ly( one is re8uired to use the 6L69P risk"reduction &rinci&le 6L69P re8uires in #oth cases that4 risk shall #e reduced so far as reasona#ly &ractica#le 6L69P does not )o#viously* say4
9isk reduction may cease -hen cost is grossly dis&or&ortional to #enefits. Mo 9%6 is im&lied 6s risk is reduced( the less it is necessary &ro&ortionately to s&end to reduce it further But #oth of these claims are in the '/% E1C0F e3&lanatory diagramN
$ %%
23 !"r# 2005
esign task
't is there#y &ossi#le that in a &articular case 6L69P -ould re8uire a further reduction in risk #eyond that set #y the S'L
$
23 !"r# 2005
%5
$ 23 !"r# 2005 %6
+iffic6lt E=a(ple
There is an e3am&le in -hich the Leveson"9isk does not e8ual the e3&ected level of loss )Ladkin 1BB;*
Mot every accident se8uence &asses through a ha+ard state There is a stochastic de&endence amongst the &ossi#le accident se8uences
$ 23 !"r# 2005 %,
$ 23 !"r# 2005 %8
something that can come to &ass( inde&endently of the severity of its harmful conse8uences a 0situation1( -hich in turn is a 0circumstance1
't seems similar to the conce&t of an accident )-hich ho-ever is an event*( #ut in -hich the severity is a#stracted a-ay
The conce&ts a&&ear to #e interdefina#le( given the #asic ontology )Ladkin( 200I* $
%1
23 !"r# 2005
The refinement of accidents into ha+ardous events and e3&licit severity may -ell #e a&&ro&riate for( say( &rocess control. /3am&le4 6 &ressure vessel #reaches )event ty&e( encom&assing many event ty&es from leaks to e3&losions* Severity4 's the #reach small or large. ,as near#y e8ui&ment heavily damaged( lightly damaged( or not at all. ,ere near#y &eo&le in5ured. Severely in5ured. ,ere some killed. 6nd ho- many of those &eo&le -ere there.
$ 50
23 !"r# 2005
Safety Functions@S9Ss4 a restricted conce&t 9isk 9eduction4 generally a good idea( #ut a&&lication is
restricted #oth in suita#ility to the a&&lication domain and statistically
J system"ty&es4 restricted( sometimes misleading conce&t S'L4 restricted and misleading 6L69P4 in &rinci&le strong( in &ractice -eak. 't strains
against &roven techni8ues such as 9isk :atri3 classification. 6 legal &rinci&le -hose technical translation is not yet clear.
$
23 !"r# 2005
51
The End
$ 23 !"r# 2005 52