Aurorisk FR

Chapter 1 · Lesson 1.1

The causal chain and its limits

Most dependability methods look for what can fail, then for what the failure causes. This works well for components, much less for systems where accidents come from interactions between elements that work.

The event chain modelInitiating cause ; Intermediate causes ; Feared event ; ConsequencesInitiating causefailure, error, deviationIntermediate causestraced back step by stepFeared eventConsequences
The image of fault trees, bow-ties and cause trees: each link is a defect. It does not see reliable components that interact badly.
Chapter objectives
  • Explain why an accident can occur without any component failing
  • Define constraint, control loop and process model (STAMP model)
  • Place the four steps of the STPA method

The chain-of-events model

An accident is often described as a sequence of events: an initial cause, intermediate causes, a top event, consequences. This is the picture behind fault trees, bow-ties and the causal tree. It assumes two things: causes can be traced back step by step, and each event is a fault (failure, error, deviation).

Assumption of the modelWhat it allowsWhat it leaves out
Accidents have a root causeTargeting the corrective action on one pointAccidents with multiple causes that are not linked to each other
Components fail randomlyCalculating probabilities (λ, PFD)Design, specification and software flaws
Component reliability makes safetyRedundancy, maintenance, testingReliable components that interact badly
Operators are a source of errorsProcedures, trainingThe context that made the operator’s action reasonable at the time

Where it falls short

Today’s systems combine software, automation, operators and organisations. Software does not wear out: if it has a flaw, it had it from the start, and it will execute it faithfully. A failure probability calculation then makes no sense, yet the risk is very real.

Reliable is not safe

A component can be perfectly reliable (it always does what it was designed to do) and still be dangerous, if what it was designed to do does not match the real situation. Failure-centred methods do not see this case.

Key takeaways
  • The causal chain describes component failures well, interactions poorly.
  • Software has no failure rate: its flaws are design errors.
  • The reliability of a component is not enough to guarantee the safety of the system.