Model-Based Development of Embedded Systems

Size: px

Start display at page:

Download "Model-Based Development of Embedded Systems"

Andrew Barton
5 years ago
Views:

Bern, 2013-07-09 Model-Based Development of Embedded Systems Challenges in System Evolution 6th Seminar Series on Advanced Techniques & Tools for Software Evolution Bern, Switzerland

1 Bern, Model-Based Development of Embedded Systems Challenges in System Evolution 6th Seminar Series on Advanced Techniques & Tools for Software Evolution Bern, Switzerland Bernhard Schätz with contributions from Sascha Kirstan, Florian Deißenböck, Benjamin Hummel, Elmar Jürgens, Stefan Wagner, Bark Albatran, fortiss GmbH An-Institut Technische Universität München

2 And now for something completely different... Why SATToSE might look at MBEES - economically and scientifically 2

3 Embedded Systems - A Hidden Software Market ca. 20 Mio. (Total users in 2012) Users ca. 2 Mio (2013 sold vehicles) 13 Bill. (2013 Turnover) Turnover 7 Bill. (10% SW-value of 2013 Turnover) 3

F:"1970"the"embedded"soPware"of"a"car"had"a"size"of"ca."100"LOC."What"is" the"average"size"2010? A:"1"Million"LOC;"premium"vehicles"up"to"10"Millionen"LOC.

4 Automotive Development: Challenges F:"How"long"is"the"harness"of"the"Phaeton"and"what"is"its"weight? A:"3.860m"and"64"kg. F:"What"is"the"increase"in"the"number"of"ECUs"of"the"previous"to"the"2010" Mercedes"SIClass? A:"By"64%"from"45"ECUs"up"to"72"ECUs. F:"1970"the"embedded"soPware"of"a"car"had"a"size"of"ca."100"LOC."What"is" the"average"size"2010? A:"1"Million"LOC;"premium"vehicles"up"to"10"Millionen"LOC. F:"What"happend"with"a"2002"USAImade"new"Beetle"in"case"of"a"broken" tail"brake"light? A:"The"automaWc"gear"was"complete"blocked. F:"If"all"warranty"costs"of"embedded"soPware"could"be"saved,"how"would" this"affect"the"ebit? A:"Increase"of"EBIT"by"ca."15I20%.

Study: Automotive MBSE What is the benefit of a model-based design of embedded software systems Manfred Broy Technical University Munich, Germany Sascha Kirstan Altran Technologies, Germany Helmut

5 Study: Automotive MBSE What is the benefit of a model-based design of embedded software systems Manfred Broy Technical University Munich, Germany Sascha Kirstan Altran Technologies, Germany Helmut Krcmar Technical University Munich, Germany Bernhard Schätz fortiss, Germany Jens Zimmermann Altran Technologies, Germany in the car industry? ABSTRACT Model-based development becomes more and more popular in the development of embedded software systems in the car industry. On the websites of tool vendors many success stories can be found, which report of efficiency gains from up to 50% in the development, high error reductions and a more rapid increase of the maturity level of developed functions (The Mathworks, 2010) (dspace 2010) just because of model-based development. Reliable and broadly spread research that analyze the status quo of model-based development and its effects on the economics are still missing. This article describes the results of a global study by Altran Technologies, the chair of software and systems engineering and the chair of Information Management of the University of Technology in Munich which examines the costs and benefits of model-based development of embedded systems in the car industry. 1. INTRODUCTION In the last 20 years the value chain in the car industry has changed drastically. All car producers and suppliers worldwide have worked on improvements in the area of mechanics, the improvement of quality requirements, and improvements in the logistic area. A lot of the potential in these areas is already exploited. A main differentiation factor turns out to be the electronics area, where a change from hardware to software development is carried out. The meaning electronics will have in the next years has been analyzed by a study of Mercer Management Consulting (Mercer, 2004). The study focuses mainly on the question how the cost factors in the development of a car will change until the year 2015 in comparison to the year In 2015 the costs for the development of electronics will have a value of 35% of the total car production costs. Whereas areas as power train and body have small increases, the costs for the development of electronic systems will be almost tripled. The predicted increases result from a variety of innovations which are being expected in this area. The majority of innovations are realized with embedded systems and especially with software. 90 percent of the future innovations in the car will be based on electronics and from that 80 percent will be realized by software" (Lederer, 2002). However, today s software development has big challenges to master like shortened development times for the cars in total versus longer development times for the software, high safety requirements and especially the growing complexity because of the rising number of functions and the increasing interaction between the functions. To master these challenges car producers and suppliers conduct a paradigm change in the software development from hand-coded to model-based development. Participants: ~180 interviewed in 67 interviews 14 Countries (u.a. D, F, I, FI, US, J, CA, BR) Profiles: Management, Developers, R/D-Members Method: Questionaire What is the benefit of a model-based design of embedded software systems in the car industry? Manfred Broy, Sascha Kirstan, Helmut Krcmer, Bernhard Schätz, Jens Zimmermann. In: Emerging Technologies for the Evolution and Maintenance of Software Models. IGI Global,

6 Automotive Modeling Languages Modeling Languages for Automotive Software: Originally modeling of functionality using control theory Primary formalism: Data flow (ASCET, Matlab/Simulink) Extension: State-based models Size: blocks 6

Automotive MBSE Simulation System Function Model

MiL-Test RCP Prototyping HW + Vehicle Production

7 Automotive MBSE Simulation System Function Model Discretation, Scaling, Augemntation, Embedding MiL-Test RCP Prototyping HW + Vehicle Production Model Autocoding Simulation System Production Code SiL-Test PiL/HiL-Test Controller + Simulation of Environment Integration Controller + Vehicle 7

8 Models: Languages and Use ML/SL/SF 92 SCADE 1,5 Statemate 4,5 29 % UML Rhapsody 6 71 % ASCET 6 Others 0 Modeled Not modeled 0% 20% 40% 60% 80% 100% Use of Models: Extensive description of (control-)functionality 8

9 Application: Functional Specification/Code Generation 50% Functional Specification Code Generation Validation/Verification 40% % % % % Not planned Planned < 1 Year 1-4 Years 5-10 Years > 10 Years 0 Use of models: Functional specification, code generation 9

10 Implementation: Code Generation 100% 80% 73,0% 77,0% 5 % 60% 40% 44,0% 95 % 20% 0% Yes No -20% Extent of Generation Average Safety Functions Others Code Generation: Substantial Extent (> 90% with 40% of participants) 10

11 Implementation: Implementation Costs/Time 0% 20% -10% -20% 0% -30% -27,0% -20% -40% -50% -60% -46,0% -48,0% -40% -60% -26,0% -47,0% -50,0% -70% -80% -80% -90% Change of Cost -100% Change of Time 50% gen. Code 51-69% gen. Code 70% gen. Code Autocoding: Implementation efforts loose significance 11

12 Overall Development: Cost Savings 0% -10% -20% -18,0% -30% -28,0% -29,0% -40% -50% -60% Intensitiy of Modeling and Testing < 25% 51% - 75% > 75% MBSE: Substantial savings, key activities in analysis/design 12

13 At a Glance: Quick Wins MBSE: Success Story : Payoff after ca. 3 Year Reduced Development Time and Costs around 30% Reduced Implementation Effort around 72% MBSE: Where is the beef? Early validation (RCP, MiL,..) discovers up to 60% of design flaws RCP amounts to 20% cost and time reduction MBSE: Quality is for free! Constructive: Problem Orientation and Complexity Reduction Analytic: Front loading of quality assurance 13

Software Evolution Issues Lehmann s Laws: Excerpt Continuing Change: A system must be continually adapted else they become progressively less

Continuing Growth: The functional capability of systems must be continually increased to maintain user satisfaction over the system lifetime.

14 Software Evolution Issues Lehmann s Laws: Excerpt Continuing Change: A system must be continually adapted else they become progressively less satisfactory in use. Continuing Growth: The functional capability of systems must be continually increased to maintain user satisfaction over the system lifetime. Increasing Complexity: As a system is evolved its complexity increases unless work is done to maintain or reduce it. Declining Quality: Unless rigorously adapted to take into account for changes in the operational environment, the quality of a system will appear to be declining.

15 Automotive Evolution Drivers Drivers of Evolution: Large base of individual, longliving configurations Ever-increasing demand for safety and comfort Supply-chain oriented production 15

16 Automotive Software and Evolution Issues Automotive Software: Examples of Evolution Issues Review-Oriented Process: Automation of Review Tasks Substantial Test Efforts: Reduction of Regression Testing Variant-Rich Software: Identification of Differences/Commonalities Functionality-Driven Design: Automation of Implementation Model-Based Development: Examples of Support Methods Automation of Review Tasks: Guideline Checker, Pattern Detectors Reduction of Regression Testing: Change Impact by Dataflow Analysis Differences/Commonalities: Model Diffs, Clone Analysis Automation of Implementation: Refactoring, Low-Level Design Pattern 16

17 What s a Clone? Software clones are segments of code that are similar according to some definition of similarity. Ira Baxter, 2002 Reusing functionality is good engineering practice Not being aware of reuse is a dangerous pitfall Clone analysis detects unwanted forms of reuse 17

18 Dataflow Clones Basic Criteria: Functionally independent (Abstracted elements) Reusable (Connected) Functionally complex (Size of functional elements) General (Number of instances) Clones detection: Find maximal, connected similar subgraphs 18

19 Clone Detection Pipeline Abstraction Simulink Parser Flat Labeled Graph Simulink Models Simulink Files Detection Clustering Visualization Clone Pairs Clone Classes 19

20 Practical Case Study Results Number of components Size of connected component Simulink/TargetLink Model (ca blocks, 71 files): Identified: 139 clone classes after filtering Most clones are relatively small & singular/infrequent Most clones affect several files/transcend several hierarchies Includes clones of library blocks 37% of relevant blocks are part of at least one clone class Clone elimination substantially reduces models 20

21 Practical Case Study Results Clone Size Number of Clone Classes > Simulink/TargetLink Model (ca blocks, 71 files): Identified: 139 clone classes after filtering Most clones are relatively small & singular/infrequent Most clones affect several files/transcend several hierarchies Includes clones of library blocks 37% of relevant blocks are part of at least one clone class Clone elimination substantially reduces models 20

22 Practical Case Study Results Cardinality Clone of Clone Size Class Number of Clone Classes > Simulink/TargetLink Model (ca blocks, 71 files): Identified: 139 clone classes after filtering Most clones are relatively small & singular/infrequent Most clones affect several files/transcend several hierarchies Includes clones of library blocks 37% of relevant blocks are part of at least one clone class Clone elimination substantially reduces models 20

23 Practical Case Study Results Cardinality Number Clone of of Clone Size Models Class Number of Clone Classes > Simulink/TargetLink Model (ca blocks, 71 files): Identified: 139 clone classes after filtering Most clones are relatively small & singular/infrequent Most clones affect several files/transcend several hierarchies Includes clones of library blocks 37% of relevant blocks are part of at least one clone class Clone elimination substantially reduces models 20

24 Practical Case Study Results Cardinality Number Clone of of Before Clone Size Models Elimination Class Number After of Clone Elimination Classes # Nodes #Edges # Nodes # Edges SIM SEM ECW AUT > Simulink/TargetLink Model (ca blocks, 71 files): Identified: 139 clone classes after filtering Most clones are relatively small & singular/infrequent Most clones affect several files/transcend several hierarchies Includes clones of library blocks 37% of relevant blocks are part of at least one clone class Clone elimination substantially reduces models 20

25 Syntactic vs Semantic Data Flow Clones 1 In 0.7 I 1.8 P 1 z I-Delay 1 Out 1 In.2 I 2 P 1 z I-Delay 1 Out Notion of Similarity Syntactic clones (Type 3): Topologically equivalent dataflow ( ) Semantic clones (Type 4): Computationally equivalent dataflow ( ) 21

26 Semantic Clone Detection Pipeline Transformation Rule Flat Labeled Graph Normalization Detection Clone Pairs 22

27 Semantic Clone Detection Pipeline Flat Labeled Graph Normalization Detection Detection Transformation Clone Pairs 22

28 Automotive Case Study Results Rule #(Execu,ons Gain"for"MulWplying"by"Constant 104 Bias"for"Adding"a"Constant 58 Joining"ConsecuWve"Product"Blocks" 42 Joining"ConsecuWve"Sum"Blocks" 40 Placing"Gain"Block"before"Integrator"Block" 28 Joining"ConsecuWve"Gain"Blocks" 16 Sum"Rule"in"IntegraWon 6 Power"Rule 5 DistribuWon"of"MulWplicaWon"over"AddiWon" 4 Replacing"Comp."to"Const."by"Comp."to"Zero 4 Placing"Gain"Block"before"DerivaWve"Block" 2 Joining"ConsecuWve"Mux"Blocks" 2 Joining"ConsecuWve"Bias"Blocks" 2 Trigonometric"funcWons" 2 EliminaWon"of"Rounding"Blocks" 2 Math"funcWons" 2 Replacing"Unary"Minus"Block"by"Gain"Block 2 Normalization of Simulink model (ca blocks): - Substantial transformation: 321 applications of 16 rules 23

29 Automotive Case Study Results Clone(Size Number(of(Clones (without(normaliza,on) Number(of(Clones (with(normaliza,on) 4I I I I I >" Total Normalization of Simulink model (ca blocks): - Substantial transformation: 321 applications of 16 rules - More clones: 87 clones vs. 68 clones 23

30 Automotive Case Study Results Clone(Size Number(of(Clone( Classes (without(normaliza,on) Number(of(Clone( Classes (with(normaliza,on) 4I I I I I >" Average(clone(size 12,7 14,5 Normalization of Simulink model (ca blocks): - Substantial transformation: 321 applications of 16 rules - More clones: 87 clones vs. 68 clones - Larger clones: 14.5 blocks vs blocks 23

31 Automotive Case Study Results Clone(Class(Cardinality Number(of(Clone(Classes (without(normaliza,on) Number(of(Clone(Classes (with(normaliza,on) Total(number(of(clones Normalization of Simulink model (ca blocks): - Substantial transformation: 321 applications of 16 rules - More clones: 87 clones vs. 68 clones - Larger clones: 14.5 blocks vs blocks - More frequent clones: 42 classes vs. 49 classes 23

32 Automotive Case Study Results Clone(Class(Cardinality Number(of(Clone(Classes (without(normaliza,on) Number(of(Clone(Classes (with(normaliza,on) Total(number(of(clones Normalization of Simulink model (ca blocks): - Substantial transformation: 321 applications of 16 rules - More clones: 87 clones vs. 68 clones - Larger clones: 14.5 blocks vs blocks - More frequent clones: 42 classes vs. 49 classes - New clones:2 additional classes 23

33 Conclusion Model-Based Evolution of Embedded Systems: Model-based development is an established best-practice development paradigm in embedded system (especially automotive) Embedded systems (especially automotive) is a heavily evolutiondriven industry Evolution in model-based development requires and enables different support mechanisms 24

34 25

Significant Reduction of Validation Efforts for Dynamic Light Functions with FMI for Multi-Domain Integration and Test Platforms

Significant Reduction of Validation Efforts for Dynamic Light Functions with FMI for Multi-Domain Integration and Test Platforms Dr. Stefan-Alexander Schneider Johannes Frimberger BMW AG, 80788 Munich,