SMS: a Spinning Mill Simulator made in Swarm

1
SMSa Spinning Mill Simulatormade in Swarm

• m.morini_at_labor-torino.it

Matteo Morini
http//www.labor-torino.it
Centre for Employment Studies LABORatorio R.
Revelli
2
Production managementthe problem

• Matching orders and production units (PU)
• Appropriate sequencing of each chain
• Who is supposed to do what AND in which sequence?

3
Production Constraints

• Timely delivery
• Soft technical constraints
• Hard technical constraints

4
Solutions evaluation

• Modelling the firms supply chain
• Simulating the production process for each order
• Accounting for
• Set-ups
• Lead times
• Technical similarity of products in a row
• Speed of heterogeneous PUs

5
Search for good production plans

• GA (G. Ferraris Genetic Manipulator)
• Each solution is coded in form of a binary string
• The SCN model can evaluate each plans fitness in
  terms of tC (minus total costs)

6
Fitness evaluation (in-depth)

• Delay costs
• Setup costs
• Misplacing order costs
• Average count costs
• Overlaying setups costs

7
Fitness evaluation (in-depth)

• Delay costs
• Setup costs
• Misplacing order costs
• Average count costs
• Overlaying setups costs

8
Fitness evaluation(Delay setup)
Cost
Total cost
Delay cost
Setup cost
Lot size
9
Late deliveries cost
Cost
Delay
10
Set-up costs
Set-up
11
Fitness evaluation (in-depth)

• Delay costs
• Setup costs
• Misplacing order costs
• Average count costs
• Overlaying setups costs

12
Misplaced orders costs

• Different yarns need different bobbins
• Cylinder-shaped, ready for dyeing
• Cone-shaped, ready for weaving
• Extremely urgent deliveries may end being made on
  the wrong kind of PU
• Re-winding a certain amount of product has a
  known cost, epxressed in money/kg.

13
Fitness evaluation (in-depth)

• Delay costs
• Setup costs
• Misplacing order costs
• Average count costs
• Overlaying setups costs

14
Average count costs
Cost
Optimal average count
Preceeding and subsequent departments shortage
Preceeding and subsquent departments saturation
Count
15
Fitness evaluation (in-depth)

• Delay costs
• Setup costs
• Misplacing order costs
• Average count costs
• Overlaying setups costs

16
Overlapping (interfering)setup costs
Cost
simultaneous setups
17
The plannerevolution and working modes

1. RANDOM planner (1)
2. Explicit rules (failed attempt) (2)
3. Optimizing by Genetic Algorithms (3)
4. Applying a human-made program (4)

18
Comparing results

• Total costs obtained on test case
• (1) random 100.00
• (3) GA (5 run) 68.75
• (3) GA (30 run) 62.25
• (3) GA (6h run) 60.62
• (4) hand made by human planner 82.27

19
Encoding sequences intobinary strings
8 1 7 2 4 3 5 1 3 2 6 3 2 1 11 2
9 3

• From base10 into base2 (using enough bits)

1000 01 0111 10 0100 11 0101 01 0011 10
0110 11...

• Semantics is lost from the GA perspective

100001011110010011010101001110011011...
20
Solutions evolution graph
21
SMS in action (w/GUI)
22
SMS batch mode sms b

• EC grant (project Penelope)
• Cordis Fifth Framework IST programme
• http//www.cordis.lu/fp5/home.html
• SMS to become part of a bigger framework
• SMS as the engine of a web-based production
  planning service

23
SMS is ERA compliant
24
m.morini_at_labor-torino.it
Matteo Morini
http//www.labor-torino.it
Centre for Employment Studies LABORatorio R.
Revelli
SMSa Spinning Mill Simulatormade in Swarm