Embed Size (px)
Citation preview
Extending the smart-card personalisation system by the graphical treatment
Angelika MaderUniversity of Twente
Cybernetix – smart cardpersonalisation system
personalisation stations
conveyer belt
unloader
printer flip
over
laser engraver
flip over
printer loader
personalisation stations
conveyer belt
unloader
printer flip
over
laser engraver
flip over
printer loader
1
1 1
1
1 1 1 1 1 11 1
1 1 1
1 11 1
1
1
Cybernetix- smart card personalisation system
way of a single card through the system
Cybernetix - smart card personalisation system
Research question: can we use model checking tools to find an optimal schedule?
optimal means optimal throughput; because the belt is not moving with constant speed, it is not the number of gaps that we optimize.
First approach: isolation of the personalisation part
Isolation of the personalisation part
personalisation stations
conveyer belt
unloader
printer flip
over
laser engraver
flip over
printer loader
processing times:Personalisation 10-50
Unloader, Loader, Flip-Over 2
Printer 3
Laser Engraver 4
Belt movement 1
dominates:
smart card personalisationsuper single mode
1
1
1
1
2
1
2
1
23
1
23
1 2
34
1 2
34
1 2
34
1 2 3
4
1
2 3
45
1
2 3
45
1
2 3
4
5
1
2 3
4
5
6
1
2 3
4
5
6
1
2 3
4
5
6
7
1
2 3 45
6
7
12
3 45
6
7
12
3 45
6
7
12
3 45 6
7
12
3 45 6
78
12
3 45 6
78
123
45 6
78
23
45 6
78
23
45 6
7
8
23
45 6
7
89
23
4
5
6
7
89
3
4
5
6
7
89
3
4
5
6
7
8
9
3
4
5
6
7
8
9
10
345
6
7
8
9
10
45
6
7
8
9
10
45
6
7 89
1011
456
7 89
1011
56
7 89
1011
56
7 89 10
11
56
7 89 10
1112
1 2
3
1 2 3
45
etc.
belt
unloader
loader
personalisation stations
First results
• problem difficult for model checking: does not scale up to great numbers of cards
• for a periodic schedule we need some sort of cycle detection
More first results
By elementary combinatorial argumentation:
the throughput of the super-single mode is k
max{4k+3,p+2}
The super-single mode meets the theoretical upper bound for throughput,if p >= 4k+1 (i.e. personalisation time is long enough w.r.t. number of stations)
k: even number of stationsp: personalisation time
Cycle length
Alternative Architecture
7
4
6
3
5
2
1
8
Why:
• allows for an easier schedule and easier analysis
• argumentation transfer to the more complicated super-single mode
• different composition properties: good for comparison
10 9
8
7
4
6
3
5
4 7 3
6
2 5
1
8 8 4 7
3
6 2
5 1
4
8
3
7
2
6
1
5
4
8
3
7
2
6
5
9 9 4
8
7
3
6
2
5
9
8
4
7
3
6
5
10 9
8
4
7
3
6
2
5
10
8
9
7
4
6
5
11 10
8
9
7
4
6
3
5
11 10
8
7
9
6
4
5
11
8
10
7
9
6
4
5
11
8
10
7
9
6
5
12 8 11 7
10
6 9
5
1212 8 11
7
10 6
9 5
8
12
7
11
6
10
5
9
Even more first results
By elementary combinatorial argumentation:
the throughput of the alternative architecture is k
max{4k,p+2}
The alternative schedule meets the theoretical upper bound for throughput,if p >= 4k-2 (i.e. personalisation time is long enough w.r.t. number of stations)
k: number of stationsp: personalisation timeCycle
length
Personalisation & graphical treatment
personalisation stations
conveyer belt
unloader
printer flip
over
laser engraver
flip over
printer loader
• cards are processed on the belt• cards do not overtake each other• graphical treatment only adds delays
processing times:Personalisation 10-50
Unloader, Loader, Flip-Over 2
Printer 3
Laser Engraver 4
Belt movement 1
Personalisation & graphical treatment
• cards leave the personalisation part with a certain delay pattern (rythm) that depends on the schedule
(the belt does not move with constant speed!)• the time pattern interferes with the delays by the graphical treatment:
extreme cases are that the graphical treatment delays synchronize completely with the time pattern of the personalisation part and have no negative effect at all, or contrary, that the delays are added completely to the production time• personalisation schedules interfere differently with graphical part;
this holds even for different optimal schedules (will be shown by experiments)• timing analysis of interference does not seem possible using elementary reasoning• use UPPAAL for throughput analysis of the composition of personalisation part and graphical part
Timing analysis - idea:
Add explicit scheduling process in the UPPAAL model,that enforces super-single mode or the alternative schedule for the personalisation part.
Personalisation & graphical treatment
process Personalisation1{
clock pers_time;
state PERSONALISING, IDLE;
init IDLE;
trans IDLE -> PERSONALISING{
guard card_id[0]==0, // no card in the personalisation station
Belt[1]>0, // there is an unpersonalised card on the belt
moving==0; // belt must stand still
sync s_p1?;
assign pers_time:=0,
card_id[0]:=Belt[1], // load card in the pers. station
Belt[1]:=0;}, // position on the belt gets empty
PERSONALISING -> IDLE{
guard pers_time>=Personalise,
Belt[1]==0, // belt cell under station is empty
moving==0; // belt must stand still
sync s_p1?;
assign Belt[1]:=-card_id[0], // put pers. card on the belt
card_id[0]:=0;}; // personalisation station is empty now
}
process Scheduler{
state S1, S2, S3, S4, S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, …, S95;
init S1;
trans
S1 -> S2{sync s_ul!; }, // synchonizes with unloader
S2 -> S3{sync s_m!;}, // synchronizes with belt
S3 -> S4{sync s_m!;}, …
S4 -> S5{sync s_ul!; },
S5 -> S6{sync s_m!;},
S6 -> S7{sync s_m!;},
…
S21 -> S22{sync s_m!;},
S22 -> S23{sync s_ul!; },
S23 -> S24{sync s_m!;},
S24 -> S25{sync s_p1!;}, // synchronizes with personalisation station 1
S25 -> S26{sync s_p2!;}, // synchronizes with personalisation station 2
S26 -> S27{sync s_p3!;}, …
S27 -> S28{sync s_p4!;}, ...
Personalisation & graphical treatment
Timing analysis – experiments:
• super-single mode and alternative schedule• 4 and 8 personalisation stations• pick&drop times (1/2 time unit or zero) at personalisation stations• time measurements for 12,16,20,24 and 16,24,32,40 cards until cycle length is determined.• personalisation times 10,20,30,40,50• cost-optimal UPPAAL
Timing analysis -experiments
number of cardspersonalisation time
12 16 20
10 113 140
137 171
161 202
20 129 148
153 179
177 210
30 167 168
199 201
231 234
40 207 198
249 241
291 284
50 247 228
299 281
351 334
4 stations & graphical treatment, super-single mode & alternative architecture
Timing analysis -experiments
number of cardspersonalisation time
12 16 20
10 113 +24 140
137 +24 +31 171
161 +31 202
20 129 +24 148
153 +24 +31 179
177 +31 210
30 167 +32 168
199 +32 +33 201
231 +33 234
40 207 +42 198
249 +42 +43 241
291 +43 284
50 247 +52 228
299 +52 +53 281
351 +53 334
4 stations & graphical treatment, super-single mode & alternative architecture
First new results
• decomposition helps to analyse more complex scheduling problems
• results from the analysis of the first part go into a explicit scheduler of the larger system (model)
• cycles could be detected (because we know batch size = cards per cycle)
• with cycle length we also have throughput
Second version of graphical treatment
cards are printed on one side only:
• Flip-overs (and laser engraver) are not in use
• each card can be printed at first OR second printer
• scheduling problem: what is the best schedule for the printers when the personalisation part is in super-single mode or the alternative schedule?
• can be solved with cost-optimal UPPAAL by similar approach as for the first version
personalisation stationsunloade
r
printer flip
over
laser engraver
flip over
printer loader
Conclusions so far• Personalisation part is dominant but not the only source of delay in the system
• Different optimal schedules can interfere differently with graphical treatment part: optimality is not
compositional
• Even if the whole problem is not (yet) solvable with model checking, model checking can be used
for parts of solutions
• Decomposition method for complex schedules can help to find good schedules
• Mixed strategies can help
What has to be done
• Schedule and timing analysis for faulty cards: performance models?
• More experiments with different architectures (e.g. numbers of cells on the belt)
