Research On The Deriviation Of AADL Design Models For Safety-Critical Software

Nowadays,MDD(Model Driven Development),especially formal MDD has gained widely concerned and has been increasingly considered as a practicable solution for ensuring safety and realiability of SCSs(Safety-Critical Softwares),and AADL is a formal modeling language standard which is widely used in safety-critical area.However,it is utmost difficult to build a AADL model manually for SCSs,especially for large-scale SCSs,due to the the diversity of elements and complexity of semantics of AADL.The(semi-)automatic deriviation of AADL design models is a widely concerned problem for academias and industries.This paper proposes a method on automatic deriviation of AADL models for safety-critical softwares,including a requirement specification method based on restricted natural language and its automatic transformation to AADL models,and a method on AADL model deriviation from source code.This paper also proposes prototypes for these proposed methods.The main contribution of this paper is given as follow:Firstly,the lifecycle of MDD seldom involves requirement,it is probably because that requirements are recorded as free textual documents,and the causes of severe accidents rised by SCSs can be traced to requirement issues,especially safety requirement issues.Therefore,to bridge the gap between requirements in natural language and MDD based on AADL,this paper proposes requirement specification method based on restricted natural language named RNLreq and its automatic transformation to AADL,named RNL2 AADL.RNLreq can reduce the ambiguity in requirement specification by structural organization of requirements,such as data dictionary,glossary,requirement templates and restricted sentence patterns,under the premise of barely changing engineers' habits.Secondly,a lot of requirements and designs only exist in historical code,which is not feedback to requirement documented by natural languages during the lifecycle of development of SCSs.Thus,it is also a notable problem to derive AADL models from source code.This paper proposes an automatic transformation from C code to AADL model,named C2 AADL,including a C code parsing with AST(Abstract Syntax Tree),and an intermediate model CAInterM,as well as transformation algorithms from C to CAInterM and from CAInterM to AADL.C2 AADL can extract the structure of codes which can reduce the workload of reuse.Finally,we realized a framework on automatic deneration of AADL models,including 3 prototypes: RNLreqTool,RNL2 AADLTool and C2 AADLTool,respectively,and validate the proposed methodologies with an industrial case study(the rocket launching system)from aerospace industry.