iteration of analysis. Therefore, MASS executes the following algorithm, introduced below as a special section.
5.1 GAFT Extension: The Method of Active System Safety
Method of active system safety assumes existence of internal knowledge about an
object in the form of dependency matrix.
• Formation of initial dependency matrix for a node using expert opinion;
• Assigning for each link between elements a direction and weight (probability);
• Formation of a list of possible recovery actions for every element;
• Set threshold ɛ to analyze and terminate dependencies;
• Set threshold q to analyze and terminate backward search;
• Repeat: During application execution, validate the program output according to
application dependent rules;
• If discrepancy is detected, then
– Use dependency matrix to build a dependency graph using the suspected
element as root node. Multiply probability along edges and stop when the
probability is smaller than ɛ;
– Form a sorted list of all scenarios according to the probability;
– Start backward search of all possible faulty elements that are predecessors of
the suspected element;
– Continue until multiplicative probabilities of dependencies are bigger than q;
– Form list of possible reasons of fault manifestation;
– Choose further actions from list of recovery;
– Use new node data and already existing data to update the dependency
matrix.
This sequence allows controlling an object in case of deviation of behavior and
recovering its condition or performing graceful degradation when full recovery is
impossible.
5.2 GAFT Derivation: A Principle of Active System Safety
This section introduces the Principle of Active System Safety (PASS) as logic
derivative of GAFT. In this approach, all key elements of a system (hardware and
software behavior) are modeled as a set of interconnected elements with dependencies reflecting the impact on each other from the recovery point of view.
Schagaev et al. [1] applied the principle of Active Safety to aviation. In their
approach, information about the proper functioning of an airplane component is
monitored in real time and compared to a list of fault symptoms.
5 GAFT Generalization: A Principle and Model of Active System…
49
5.1 GAFT Extension: The Method of Active System Safety
Method of active system safety assumes existence of internal knowledge about an
object in the form of dependency matrix.
• Formation of initial dependency matrix for a node using expert opinion;
• Assigning for each link between elements a direction and weight (probability);
• Formation of a list of possible recovery actions for every element;
• Set threshold ɛ to analyze and terminate dependencies;
• Set threshold q to analyze and terminate backward search;
• Repeat: During application execution, validate the program output according to
application dependent rules;
• If discrepancy is detected, then
– Use dependency matrix to build a dependency graph using the suspected
element as root node. Multiply probability along edges and stop when the
probability is smaller than ɛ;
– Form a sorted list of all scenarios according to the probability;
– Start backward search of all possible faulty elements that are predecessors of
the suspected element;
– Continue until multiplicative probabilities of dependencies are bigger than q;
– Form list of possible reasons of fault manifestation;
– Choose further actions from list of recovery;
– Use new node data and already existing data to update the dependency
matrix.
This sequence allows controlling an object in case of deviation of behavior and
recovering its condition or performing graceful degradation when full recovery is
impossible.
5.2 GAFT Derivation: A Principle of Active System Safety
This section introduces the Principle of Active System Safety (PASS) as logic
derivative of GAFT. In this approach, all key elements of a system (hardware and
software behavior) are modeled as a set of interconnected elements with dependencies reflecting the impact on each other from the recovery point of view.
Schagaev et al. [1] applied the principle of Active Safety to aviation. In their
approach, information about the proper functioning of an airplane component is
monitored in real time and compared to a list of fault symptoms.
5 GAFT Generalization: A Principle and Model of Active System…
49