April 30, 2008LFM 2008 1 From Informal Safety-Critical Requirements to Property-Driven Formal Validation Alessandro Cimatti, Marco Roveri, Angelo Susi,

April 30, 2008 LFM 2008 1

From Informal Safety-Critical Requirements

to Property-Driven Formal Validation

Alessandro Cimatti, Marco Roveri, Angelo Susi, Stefano Tonetta

Fondazione Bruno Kessler

Istituto per la Ricerca Scientifica e Tecnologica

April 30, 2008 LFM 2008 2

Outline

• Requirements Analysis– Requirements engineering– Requirements specification– Requirements formal validation

• Goals of the project• The proposed methodology

– Informal analysis– Formalization– Formal validation

• Conclusions and future directions

April 30, 2008 LFM 2008 3

Requirement Analysis

April 30, 2008 LFM 2008 4

Requirements definition

• Set of properties that describe the system under development– Configurations

• Entities• Relationships

– Behaviors• What is admissible• What must not happen

• Requirements analysis is a pervasive problem in nowadays industry

April 30, 2008 LFM 2008 5

Requirements engineering

• Definitions of– Goals– Scenarios– Models– Misuse cases

• Analysis mainly based on inspection• Important for analysis of functional, quality, and

security requirements• High cost of flaws in the safety-critical specification

– Formal methods must be integrated

April 30, 2008 LFM 2008 6

Requirements Specification

• Natural language is widely used as specification language– ambiguous– hard to process automatically

• statistical natural language processing• controlled natural language

– requires background information• knowledge representation, taxonomies

• Formal languages are often too difficult• Important modeling issues

• modeling real-time and space– not only discrete data– hybrid automata as semantic model

• modeling class/instance relationship• modeling temporal evolution

April 30, 2008 LFM 2008 7

Formal languages for requirements• Logic is usually used as language for property

specification– Examples: FOL, LTL, CTL, DL, …

• In hardware design, standards for languages to represent of properties and design intent are emerging (e.g. PSL, SVA)

• Requirement → formula is usually a one-to-one mapping

• Every property/formula defines what is admissible• Opposite to model specification

– System represented by abstract machines– Examples: Automata, Petri Nets, Z, B, VDM, CCS, …

• Notably: SCR – used for requirement analysis– the set of formulas define an abstract machine

April 30, 2008 LFM 2008 8

Formal Verification Methods

• Theorem Proving– general and expressive language– both system and properties expressed in the same language– not automatic

• Model Checking– more restricted language

• system as finite state automaton• properties as temporal logic formulae

– more effective engine• based on search

– complete answer• yes, system is correct• no, here is a counterexample

April 30, 2008 LFM 2008 9

Requirements formal validation• Standard verification setting:

– the design is under analysis– the requirements are taken as “golden”– verification means checking compliance

• system has no behavior violating requirements• New focus: requirements, not model• There is no model (yet) • Here the goal is to enhance quality of requirements • No clear definition of correctness• Several ways in which requirements might be flawed

– inconsistent (contradictory)– too strict (some desired behaviors are not possible)– too weak (some undesired behaviors are admissible)– unrealizable (system would have to foresee the future to be

implemented)

April 30, 2008 LFM 2008 10

Goals of the project

April 30, 2008 LFM 2008 11

Methodology goals

• Validation of natural language requirements• Bridging the gap between natural language

and formal analysis• Enhancing the quality of the requirements• Increasing the confidence in their correctness

April 30, 2008 LFM 2008 12

Methodology Desiderata

• Usability– Avoid intricate formalisms– Hide formal methods with semiformal representations– Formalization close to the original specification

• Traceability– Link between one requirement and formal counterpart

• Automation of the verification process• Validation applicable to parts of the specification

– Hierarchical approach– Modular approach– Incremental approach

• What-if analysis

April 30, 2008 LFM 2008 13

Main issues

• Human factor in formalization• Expressiveness of the formal language• Completeness of validation• Quality of feedback• Complexity of the verification

April 30, 2008 LFM 2008 14

ETCS case study

• Specification for the interoperability between train on-board and track-side systems.

• Complex, huge set of requirements• Very ambiguous natural language

– NL cannot be substituted• Mixed components functionalities

– Global properties• Optional functionalities of components

April 30, 2008 LFM 2008 15

New Methodology

April 30, 2008 LFM 2008 16

Overview

• Step 1: Informal analysis• Step 2: Formalization• Step 3: Formal analysis

April 30, 2008 LFM 2008 17

Step 1: Informal Analysis• Supported by standard RE tools• Standard checklist-based analysis to support the formalization

– Requirements Identification– Requirements Categorization

• Customized taxonomy– Definitions– Properties (invariants, restrictions)– Functionalities– Procedures– Scenarios

– Requirements Structure• Define relationships among requirements

– Dependency links (A → B)– Conflict links (A → not B)

– Requirements Quality• Relevant, verifiable, …

– Requirements Status• Formalized, validated, to be disambiguated, …

April 30, 2008 LFM 2008 18

Step 2: Formalization

• Unified Modeling Language– de facto standard in model based design– common in software engineering– wide tool support– (semi-)formal

• very rich• unclear semantics

• We propose a restricted use of UML– clear semantics– different diagrams depending on classifications of natural language

• We extend UML with Controlled Natural Language– Highly-restricted syntax– Express invariants and temporal behaviors

April 30, 2008 LFM 2008 19

Proposed Specification Language• Class diagrams (glossary/ontology)

– Classes– Attributes– Associations– Inheritance relationships– Methods

• State machines (procedures)– States– Transitions (event [guard]/transition)

• Sequence diagrams (scenarios)– Timeline– Methods calls

• Constraints (assumptions on system and environment behaviors) – Property specification language– Predicates over the UML model– First-order terms– Quantifiers over the class universe

April 30, 2008 LFM 2008 20

Examples – class

“For each section composing the MA the following information shall be given;

a) Length of the sectionb) Optionally, Section time-out value and distance from

beginning of section to Section Time-out stop location”

Movement authority

Section

LengthDistance

….

….

INVAR Distance=Beginning_location-

Timeout_stop_location1..n

TimeoutTimeout_stop_location

Beginning_location

April 30, 2008 LFM 2008 21

Example – State Machine

• “Each procedure is defined by a set of mandatory requirements and/or by a state transition chart (where convenient).”

April 30, 2008 LFM 2008 22

Example - Constraints

• “A Balise Group is linked when its linking information is known in advance”

Onboard Balise_Group

Linked

Linking_info

….

….

BEHAVIOR always (if onboard.Receive_linking_info=linking_info

then linked=true until Location=onboard.position)

Location

LocationReceived_linking_info

April 30, 2008 LFM 2008 23

Step 3: Formal Analysis• Main checks:

– Consistency– Scenarios– Properties– Realizability

• Consistency checking– Is there a configuration and an evolution thereof consistent with the

specification?• Scenario checking

– Is there a configuration and an evolution thereof consistent with the specification and a given scenario?

• Property checking– Is there a configuration and an evolution thereof consistent with the

specification and a violation of the property?• Realizability

– Is there an implementation that realizes the specification?

April 30, 2008 LFM 2008 24

Verification procedure• Definition of the verification domain

– User defined– Finite recursion– Small domain encoding

• Symbolic encoding– Semantic model: fair infinite-state transition systems with

temporal constraints– Space minimization based on invariants

• Abstraction-based verification (CEGAR)– Automatic abstraction with SMT and BDD– Model checking with BDDs or SAT– Automatic refinement with interpolation– Tools: NuSMV, MathSAT

• Boolean abstraction of temporal formulas

April 30, 2008 LFM 2008 25

Step 3: Feedback

• Traces– Witnesses in case of consistency– Counterexample in case of property violation

• Unsat cores– Minimal inconsistent subset

• Vacuity– Irrelevant sub-formulas of the property

• Presentation of results– Sequence of method calls (sequence diagrams)– Tree of event combinations (fault-tree analysis)

26

Conclusions and future work• The methodology accomplishes the goal

– Usability– Traceability– Automation– Feedback

• Ongoing feasibility study• Ongoing development of tool support

– RSA plug-in to interact with ReqPro– RSA plug-in to support formalization– RSA plug-in with NuSMV back-end

• Many open research directions– Controlled natural language– Automatic instanciation– Infinite domain– Generalize CEGAR

April 30, 2008 LFM 2008 27

References

• Requirements validation– Alessandro Cimatti, Marco Roveri, Viktor Schuppan, Stefano Tonetta: Boolean Abstraction for Temporal Logic

Satisfiability. CAV 2007– Alessandro Cimatti, Marco Roveri, Stefano Tonetta: Syntactic Optimizations for PSL Verification. TACAS 2007– Ariel Fuxman, Lin Liu, John Mylopoulos, Marco Roveri, Paolo Traverso: Specifying and analyzing early requirements in

Tropos. Requir. Eng. 9(2): 132-150 (2004) – Ingo Pill, Simone Semprini, Roberto Cavada, Marco Roveri, Roderick Bloem, Alessandro Cimatti: Formal analysis of

hardware requirements. DAC 2006• Realizability

– Nir Piterman, Amir Pnueli, Yaniv Sa'ar: Synthesis of Reactive(1) Designs. VMCAI 2006– Alessandro Cimatti, Marco Roveri, Viktor Schuppan, Andrei Tchaltsev: Diagnostic Information for Realizability. VMCAI

2007• Fault-tree analysis

– Richard Banach, Marco Bozzano: Retrenchment, and the Generation of Fault Trees for Static, Dynamic and Cyclic Systems. SAFECOMP 2006

– Piergiorgio Bertoli, Marco Bozzano, Alessandro Cimatti: A Symbolic Model Checking Framework for Safety Analysis, Diagnosis, and Synthesis. MoChArt 2006

– Marco Bozzano, Adolfo Villafiorita: The FSAP/NuSMV-SA Safety Analysis Platform. STTT 9(1): 5-24 (2007)• Vacuity

– Ilan Beer, Shoham Ben-David, Cindy Eisner, Yoav Rodeh: Efficient Detection of Vacuity in Temporal Model Checking. Formal Methods in System Design 18(2): 141-163 (2001)

– Doron Bustan, A. Flaisher, Orna Grumberg, Orna Kupferman, Moshe Vardi: Regular Vacuity. CHARME 2005• Controlled Natural Language

– Schwitter: English as a Formal Specification Language. DEXA 2002– Nelken, Francez: Automatic Translation of Natural Language Specification Systems into Temporal Logic. CAV 1996