Define and propose modidfications to RAAML 1.1 to include reliability block diagrams
We suggest to add a new method for risk management for medical devices according to ISO 14971.
In the context of the medicussy project we have created a profile and a library. These could be used as a starting point. The files can be found in a github-repository: https://github.com/oose/medicussy. There you also find further example models that illustrate the application of the method.
We invite everybody to contribute to the github repository.
In Enterprise Architect 15.2 the CoreRAAML.xmi is not loaded completely. The generalization relation from Situation to SysML Block is not loaded. It seems to me that the xmi export has tool-specific content.
RAAML claims to implement STPA, but some important elements are missing.
For example,
In Step-1 of the STPA process (in the 2018 Leveson/Thomas STPA Handbook – Figure 2.3) it calls out the need to derive “Safety Constraints” from the Losses. I can find no reference to Safety Constraints in the RAAML.
Furthermore the RAAML does not provide any linkage from Losses to Hazards, which is expected in the Handbook.
Many parts of RAAML are broken in to a Library and Profile. The Profile contains the Stereotypes, for example the definition of the <<AND>> stereotype for an AND gate. The Library contains a definition, for example that an AND Block with the <<AND>> stereotype is a type of Gate. In other words, RAAML often has two elements in the model in different namespaces but with the same name.
In the Gaphor tool, we autocode the model in to a Python datamodel. So far while implementing UML and SysML, we didn't have to name each Python class with the full namespace because the element names were always unique. With RAAML this is no longer the case, there are often two elements with the same name.
I think this can also cause confusion for users by duplicating names. For example, if someone is talking about the AND element, are they referring to the the stereotype or the block?
SysML v2 is starting to use two similar names for the definition and the usage, but the definition includes "def" in the name. For example, the definition of a part is a "part def" and the usage is a "part". I would recommend we use something similar for RAAML. the definition of the AND gate could be "AND_Def", and the stereotype could be "AND".
The duplicated names include:
Gate, AND, BasicEvent, Cause, ConditionalEvent, DormantEvent, Early, FMEAItem, FailureMode, HouseEvent, INHIBIT, Late, MAJORITY_VOTE, NOT, OR, SEQ, TopEvent, UnsafeControlAction, XOR, ZeroEvent.
An instance of a part can be part of exactly one composite instance, but there could be many options for the composite. For example the same ScenarioStep could be part of many Scenarios. Of course, each instance of the ScenarioStep will occur only in one Scenario instance.
In order to allow this, the multiplicity on the whole side of the composite relationship in figure 9.22 must be 0..1.
In this figure this multiplicity is not shown. This leads to two possible interpretations. According to UML the default is 1, but according to SysML the default is 0..1. Now, the General Concepts Library is a SysML model, hence we could assume that it is 0..1.
However, this default is just a notational thing: "SysML: These multiplicities may be assumed if not shown on a diagram.". In order to make it into the xmi-file the multiplicities must be set explicitely. And in the current xmi-file the multiplicity is not defined. When it is opened by Cameo, it sets the value explicitely to 1. You might consider this a bug in Cameo, but I guess this behavior is within the specification.
So my question is: What was the intention? Should a scenario step only be part of one scenario? Then the multiplicity must be 1 and this should be made clear in the diagrams. Or can it be part of many scenarios? Then the multiplicity must be 0..1 and this should also be shown in the diagrams. And it should be defined in the xmi-file.
The same question needs to be answered for all composite associations in the libraries, e.g. Risk to Effect, Event and HarmPotential. ScenarioStep is just one example for the issue.
o Table 9.1 lists «ControllingAction» as “A stereotyped UML dependency” and states in the preceding text that “Situations can be mitigated, detected, and prevented via the ControllingAction. The use of this relationship introduces new safety requirements.” This term does not appear anywhere else.
o The Core Profile does, however, describe «ControllingMeasure» as “A measure taken to address (mitigate severity, reduce probability of occurrence, increase probability of detection) a potential or real adverse situation.” which is similar, but not identical to «ControllingAction»
Considering its significance, Situation is defined rather ambiguously in Table 5.1: “A situation describes a set of situation occurrences of some type. The system, place, time and state parameters are described by classifiers rather than individual descriptions. A situation occurrence is a system being in a given place at given time and in a given state.” Section 9.1 adds that “The system, place, time and state parameters are described by classifiers rather than individual descriptions.” This prompts a number of questions, which are unanswered by the specification:
• Why is there no formal definition of ‘situation occurrence’ in the Core Profile? (it is only informally defined as part of the definition of ‘situation’) The lack of a formal definition seems to be something of an oversight, given their central importance, especially considering that the stated aim of RAAML to “avoid inconsistent model-based solutions [that] prohibit direct model sharing between organizations and across the various tools”.
o My recommendation would be to add this to the Core Concepts e.g. “Figure 9.1 – Core concepts domain model”, a definition in Section 9.1.x and an entry in the 5.1 definitions table.
• In the absence of the above definition of ‘situation occurrence’, the RAAML user is left to guess how ‘situation occurrences’ should be represented in the model?
o One might initially assume that a ‘situation occurrence’ is intended to be an instances of ‘situation’, but this assumption would appear to contradict the definition of ‘situation’, which states it is a set of ‘situation occurrences’. In other words, every instance of ‘situation’ must be associated with a set corresponding instances of ‘situation occurrence’
o What ‘types’ of ‘situation occurrence’ exist (definition of ‘situation’ states “…situation occurrences of some type”)
o What is the nature of the set of ‘situation occurrences’ that are described by a ‘situation’?
Should this be represented in UML with shared or composite aggregation from whatever represents ‘situation occurrence’ to ‘situation’ – maybe not, because ‘situation’ only “describes a set”, so perhaps this should be a plain association labelled ‘describes’?
If the relationship is aggregation, should/could a ‘situation’ comprise anything else alongside the set (such as a description of the relationships between the members of the set)?
o Regardless of whether ‘situation’ aggregates ‘situation occurrences’ or merely describes them, what range of multiplicities is valid (e.g. can a ‘situation’ be an empty set, have only a single constituent ‘situation occurrence’, or must it be comprise two or more)?
• RAAML deliberately relies on UML classifiers to represent “system, place, time and state”.
o State is a UML type and RAAML shows e.g. «FailureState» extending the «State» metaclass, but which classifiers are intended to be used for ‘system’, ‘time’, and ‘place’?
o I am also left wondering why RAAML relies on classifiers for this purpose, rather than making them explicit in the Core Concepts – no rationale for this decision is provided. This seems almost certain to result in various interpretations in different organizations' tools, thus violating the stated aim "avoid inconsistent model-based solutions"
The Automotive Industry Action Group and the "Verband der deutschen Automobilindustrie" published a FMEA handbook in 2019 where they discourage the use of the Risk Priority Number (RPN) and instead suggest the use of an Action Priority to determine order and importance of identified failure modes and their effects.
There should be an explanation of what the previousRPNValues attribute means.
The section focusses on probability as an attribute from FTA elements.
This is arguably the least important attribute from FTA, particularly in system modeling and given the simplistic treatment of probability in this section.
A far more valuable attribute from FTA is the cut set, or minimal cut set at any gate within the FTA hierarchy.
This is a major shortfall that I believes renders the use of these profiles within system modeling very limited, particularly in early system work before the base event probabilities can be known. Even in later lifecycle phases, the simplistic treatment of probabilities that ignores real world interdependence renders the results from such modeling misleading at best.
How are risk matrices supported that visually relate risks to the probability and severity of their occurrence as well as to the defined risk acceptance criteria levels (e.g. acceptable/tolerable/intolerable)?
The GSN Community standard and metamodel, as of version 2 of GSN, are based on extending OMG’s Structured Assurance Case Metamodel (SACM) version 2. In a sense, all GSN elements are sub-classes of elements defined in SACM so that tools supporting GSN can leverage the SACM metamodel for exchanging GSN content. The material in RAAML that discusses GSN and define RAAML GSN constructs does not reflect this approach and appears at odds to the GSN community standard and its use and leveraging of OMG’s SACM standard.
How is issue/task/responsibility tracking supported for an (unsatisfied) ISO 26262 safety case?
