Validation of Automotive Control Applications using Formal Methods and metamodeling techniques ❖ Simone Silvetti, Esteco Spa & University Udine ❖ Mariapia Marchi, Esteco Spa
www.caeconference.com MDB ( M odel B ased D evelopment) ❖ process aimed at designing complex systems ❖ cost reduction ❖ reduce development time 2
www.caeconference.com MDB ( M odel B ased D evelopment) International CAE Conference 17 - 18 October 2016 3
www.caeconference.com MDB ( M odel B ased D evelopment) International CAE Conference 17 - 18 October 2016 4
www.caeconference.com MDB ( M odel B ased D evelopment) International CAE Conference 17 - 18 October 2016 5
www.caeconference.com MDB ( M odel B ased D evelopment) International CAE Conference 17 - 18 October 2016 6
www.caeconference.com MDB ( M odel B ased D evelopment) International CAE Conference 17 - 18 October 2016 7
www.caeconference.com MDB ( M odel B ased D evelopment) International CAE Conference 17 - 18 October 2016 8
www.caeconference.com Validation Process International CAE Conference 17 - 18 October 2016 9
www.caeconference.com Validation Process ❖ Use of block diagram tools (Simulink, Gt suite) Powerful Tools but ❖ complex International CAE Conference 17 - 18 October 2016 10 10
www.caeconference.com Validation Process ❖ Use of block diagram Use of natural languages ❖ tools (Simulink, Gt Involves time events... ❖ suite) Powerful Tools but Not rigorous ❖ ❖ complex Not Machine interpretable ❖ International CAE Conference 17 - 18 October 2016 11 11
www.caeconference.com Validation Process ❖ Use of block diagram Use of natural languages ❖ tools (Simulink, Gt Involves time events... ❖ suite) Powerful Tools but Not rigurous ❖ ❖ complex Not Machine interpretable ❖ International CAE Conference 17 - 18 October 2016 12 12
www.caeconference.com Validation Process ❖ Use of block diagram Use of natural languages ❖ tools (Simulink, Gt Involves time events... ❖ suite) Powerful Tools but Not rigurous ❖ ❖ complex Not Machine interpretable ❖ ✓ FORMAL METHODS ! International CAE Conference 17 - 18 October 2016 13 13
www.caeconference.com Validation Process International CAE Conference 17 - 18 October 2016 14 14
www.caeconference.com Validation Process International CAE Conference 17 - 18 October 2016 15 15
www.caeconference.com Validation Process φ International CAE Conference 17 - 18 October 2016 16 16
www.caeconference.com Validation Process φ International CAE Conference 17 - 18 October 2016 17 17
www.caeconference.com Validation Process φ “If the engine speed (w) is always less than k 1 then vehicle speed (v) can not exceed k 2 in less than T sec” ᅟᅠᆨ ( F [0,T] (v ≥ k 2 ) ⋀ G (w ≤ k 1 )) International CAE Conference 17 - 18 October 2016 18 18
www.caeconference.com Robustness Semantics φ ⊧ ? International CAE Conference 17 - 18 October 2016 19 19
www.caeconference.com Robustness Semantics φ ⊧ ? F ( f>k ) Boolean yes/no k International CAE Conference 17 - 18 October 2016 20 20
www.caeconference.com Robustness Semantics φ ⊧ ? F ( f>k ) Boolean Robustness +30 yes/no +30 / -30 k More Information! International CAE Conference 17 - 18 October 2016 21 21
www.caeconference.com The goal f M M(f) International CAE Conference 17 - 18 October 2016 22 22
www.caeconference.com The goal The optimization Problem min [M(f), φ ] R = f ∈ F f M M(f) International CAE Conference 17 - 18 October 2016 23 23
www.caeconference.com The goal The optimization Problem min [M(f), φ ] R = f ∈ F ≤ 0 Counterexample R ≥ 0 Safe! f M M(f) International CAE Conference 17 - 18 October 2016 24 24
www.caeconference.com The optimization process
www.caeconference.com The optimization process Challenges ❖ Low number of model execution ❖ Inputs are functions (temporal series)!! 26
www.caeconference.com The optimization process Challenges ❖ Low number of model execution ❖ Inputs are functions (temporal series)!! 27
www.caeconference.com The optimization process Challenges ❖ Low number of model execution GP-UCB ❖ Inputs are functions (temporal Adaptive Control Point series)!! Parametrization 28
www.caeconference.com The Control Point Parametrization interpolation Fix the times 29
www.caeconference.com The Control Point Parametrization interpolation Fix the times n Control Points n Variable to optimize 30
www.caeconference.com The Control Point Parametrization interpolation Fix the times n Control Points n Variable to optimize 31
www.caeconference.com The adaptive Control Point Param. interpolation n Control Points 2n Variable to optimize 32
www.caeconference.com Problem Increase the expressivity but... Doubled the variables 33
www.caeconference.com Problem Increase the expressivity but... Doubled the variables Solution GP-UCB Optimizer 34
www.caeconference.com GP-UCB International CAE Conference 17 - 18 October 2016 35
www.caeconference.com GP-UCB International CAE Conference 17 - 18 October 2016 36
www.caeconference.com GP-UCB International CAE Conference 17 - 18 October 2016 37
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 38
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 39
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 40
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 41
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 42
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 43
www.caeconference.com GP-UCB P(x,y) International CAE Conference 17 - 18 October 2016 44
www.caeconference.com Doubled the variables Reduce Input Space 45
www.caeconference.com Schema ✓ N Rob. ? GP - UCB N++ International CAE Conference 17 - 18 October 2016 46
www.caeconference.com Adaptive Idea Input Space 47
www.caeconference.com Adaptive Idea 1 Input Space 48
www.caeconference.com Adaptive Idea 2 Input Space 49
www.caeconference.com Adaptive Idea 2 Input Space 50
www.caeconference.com Adaptive Idea 2 Input Space 51
www.caeconference.com Adaptive Idea 3 Input Space 52
www.caeconference.com Adaptive Idea 3 Input Space 53
www.caeconference.com Adaptive Idea 4 Input Space 54
www.caeconference.com Adaptive Idea Input Space 55
www.caeconference.com Automatic transmission 56
www.caeconference.com Automatic transmission 57
www.caeconference.com Automatic transmission 69 blocks : 2 integrators, 3 look-up tables, 3 2D look-up tables, Stateflow Chart 58
www.caeconference.com Results International CAE Conference 17 - 18 October 2016 59
www.caeconference.com Results ✓ aCPP reduces minimum number of evaluations by 50-70% GP-UCB is slow. International CAE Conference 17 - 18 October 2016 60
www.caeconference.com Results Time = {#Simulations} x {Simulation Time} + {Optimizer time} GP-UCB is slow International CAE Conference 17 - 18 October 2016 61
www.caeconference.com Results Time = {#Simulations} x {Simulation Time} + {Optimizer time} GP-UCB is slow Future work ❖ from Matlab to Java (parallelization) ❖ multi-objective approach ❖ using fmi as simulator International CAE Conference 17 - 18 October 2016 62
www.caeconference.com Acknowledges Esteco Luca Bortolussi Alberto Policriti International CAE Conference 17 - 18 October 2016 63
www.caeconference.com ….and use Formal Methods 64
Recommend
More recommend
