Professional Documents
Culture Documents
PBS Optimization Strategy Presentation 1665132863
PBS Optimization Strategy Presentation 1665132863
PBS Optimization Strategy Presentation 1665132863
Redesigning processes
and layout
PBS Optimization
Professors: Pedro Amorim, Sérgio Castro
1M.EGI01
Sonae MC wants to redesign its PBS operation, looking for process redefinition and
a layout suggestion in order to reduce overall costs
Leading company in the food Network with +1.300 stores PBS and PBL
retail sector in Portugal strategies
Design PBS operation from scratch Inbound & Picking & Dispatch & Despicking
Define picking and dispatch strategies Verification Replenishment Transportation
Minimize total costs
Keep service level to stores Not enough information Challenge scope
Out of scope
Source: https://sonaemc.com/sobre-nos/ 2
To tackle the challenge, a 5-step process was designed with strict deadlines to
assure group consistency in actions and a quality output
Analysis and
Deconstruction Baseline Scenarization Optimization
evaluation
3
Deconstructing the problem is the first step, which involves understanding the
main drivers of cost and the necessary inputs and outputs
Analysis and
Deconstruction Baseline Scenarization Optimization
evaluation
4
Total costs can be minimized by decreasing picking, replenishment, warehousing,
transportation and despicking costs whilst keeping the same service level
Problem variables
Dispatch zone m² ↓
5
Picking time is influenced by 4 variables, being distance travelled and number of
interruptions on the weight sequence the ones with bigger room for improvement
Problem variables
Time to visit a
Setup time per pallet 3’ 25s Picker Speed 6km/h Time per interruption 1’
picking position
6
Forklift cost per time unit is bigger, but it is predictable that the overall costs will be
less significant since the number of replenishments is relatively low
Problem variables
Replenishment by Replenishment by
Moving time Up/down time
box time pallet time
Nº boxes Nº pallets
Distance ↓ Nº up/down moves ↓ ↓ ↓
replenishments replenishments
7
Warehousing costs feature two main tackling areas - storage and dispatch zone -
which must be reduced to meet our goal
Problem variables
Machine zone m²
Warehousing
Costs
Storage zone m² ↓
Dispatch zone m² ↓
1 set of lines
VS.
8
Despicking costs are a direct result of warehouse operations and despite not being
an in-warehouse cost, must be taken into consideration for SONAE’s cost analysis
Problem variables
1 Deletes the repeated successive numbers from the pallets products BU’s sequence
If there is only 1 BU
Pallet mixing Seconds per box
1 2 1 3 1 2 4 2 5
1 BU -
Checks which is the 1st number and, from end to start, if that number repeats itself
2
2 or more BU’s 1 2 1 3 1 2 4 2 5
1,5
but divided
Despicking time
Creates a list with the different BUs that appear in that sequence
9
After understanding which variables to optimize, a base scenario must be created
to understand where SONAE stands and the biggest improvement opportunities
Analysis and
Deconstruction Baseline Scenarization Optimization
evaluation
10
Inbound activities were disregarded, 0 demanded SKU's were not eliminated and a
+20% capacity dispatch area was assumed due to insufficient data
Scope Forklifts
SKU Pallets
11
Available data had to be treated, combined and analyzed in order to have the
necessary inputs to create the base scenario
SKU_data List of SKUs, containing its SKU number, Using Excel: • Add a column with the nº of times
volume, weight and BU appearing in the picking orders
• Add a column with nº of boxes that fit in a
1.5m3 pallet taking in account the box • Add a column with product classification
volume based on its frequency in picking orders1
• Add a column with nº of boxes that fits in a
1000kg pallet, considering the box weight
• Add a column with the minimum number
from the previously added columns – this will
be the final number of how many boxes fit
in a pallet
PBS_Picking List of store orders, containing the store Using Excel: Using Python:
data number, order date, SKU and number of
boxes of the ordered product • Add columns with the properties of the order • Add a shipping day, hour and minute to
SKU based on the previously mentioned each order (next shipping date after order
Excel, namely the volume, weight, BU and date)
product group
1Detailed next
12
Python was used to generate picking orders and positions- necessary inputs for the
Anylogic simulation - which outputted the variables needed to calculate costs
Treated Treated Treated Simulation
PBS_Picking data POs and PPs Despicking
PBS_Shipping data data
Python
Demanding computational effort - Excel Excel
Picking Orders:
Computes:
• Orders PO’s by shipping date, hour, Warehouse working simulation1
minute and store • Necessary dispatch area
• Assigns to each PO a pallet number accounting with 20% increase in
considering that the maximum pallet Python maximum pallets in the dispatch
volume is 1.5m³ area at the same time
Computes the despicking time based
Picking Positions: on each pallet composition2 • Costs of each activity
• Defines a WH picking position for
each SKU (storage, rack, bay, shelf) Demanding computational effort for • Total costs
• Defines the stock pallets positions Excel.
Summarizes and compares all
(above each SKU position) Independency from Anylogic, as it is modeled scenarios with the base
Other necessary treatments depending not dependent on the actual routes scenario
on the scenario to be tested
13
AnyLogic was used as the preferred simulation environment due to its visual
comprehension and route optimization, as well as facilitated error identification
AnyLogic Simulation
Why AnyLogic? Simulation Agents
Picking Flow
Order Entry Move Stock Verification Move Weight Validation
A pallet order (set of The picker goes to Picker validates if there is Picker goes to If the product is heavier
picking orders with the the 1st order SKU enough stock of SKU for the next order SKU than the previous one,
same pallet number) is picking position and next order. If not, sends a picking position the picker reorganizes
seized by a picker picks it replenishment request and picks it the pallet
Replenishment Flow
Simulation outputs
Order Entry Move
Pallet that Forklifter goes to pallet position,
needs to be brings pallet down (2min) and
replenished waits for the replenishment Total Total Nº pallets Nº pallets Max pallets in dispatch
enters flowchart (3min/box or 1min/pallets)¹ picking time forklift time shipped done off time simultaneously²
¹If it is a box, every two times, the pallet needs to go up so, to simplify the system, 1 extra minute was considered in each box replenishment
²Afterwards, this variable will be used to conclude what is the necessary dispatch area
14
In the base scenario, picking, warehousing and transportation account for 93% of
total costs, with picking, warehousing and service-level at sub-optimal values
Base Scenario Main Assumptions Base Scenario Cost Distribution and Results
6k € 195k €
8k €
POs and PPs are randomly sorted 48k €
Major improvement
Picking Warehouse Service-Level
areas:
15
After understanding the biggest improvement points, different scenarios were
tested to understand feature impact and structural opportunities
Analysis and
Deconstruction Baseline Scenarization Optimization
evaluation
16
Different features were tested individually on the base scenario to understand their
impact on specific and overall costs as well as assess possible synergies
Total Picking Forklift Despicking Warehouse Transport
100 000,00 €
210 000,00 €
80 000,00 €
190 000,00 €
60 000,00 €
170 000,00 €
40 000,00 €
- € 130 000,00 €
Base Weight SKU BU Demand Boxes 2 levels 2 1 BU per PO 24h
Scenario POs+PPs POs+PPs POs+PPs POs+PPs PPs PPs passages Pallet before ship
Picking Order and Position Sorting Picking by Boxes Picking positions in 2 levels
Main takeaways
Sorting PO and PP by the same criteria diminishes picking Increases picking time and decreases Increases picking and forklift time
time because picker will follow an organised sequence warehouse costs and decreases warehouse costs
Demand and weight showed the most impact Great for low demand SKUs Optimal for low/0 demand SKUs
17
Disruptive scenarios were tested, but didn’t find great results and added more
complexity to day-to-day operations
C+D SKU’s bulk picking and dynamic storaging I-shape, L-shape and U-shape scenarios2
Pickers wouldn’t need to go to more distant storage I-shape required different inbound/dispatch areas,
units and would pick in the end from the pile which would require more space
Requires a special area dedicated to dynamic L-shape and U-shape would have more distant
storaging storages, increasing picking costs
18
Warehouse innovative solutions were analyzed, nevertheless some of them would
not be a good investment for Sonae MC
Reduce picking time as each picker Increase in WH capacity Time and space saved in pallet
could fulfill a pallet with only one Elimination of individual access handling
path along the corridor to aisles Good for FIFO methodology
All of these options could be interesting solutions in other companies. However, the volume of Sonae MC daily orders
require a SKU storage solution and unique workers movementthat are not compatible with the technologies presented.
¹To simplify, the value only relates to the shelves length. Clearly, the warehouse structure would have to be longer due to the building.
Source: Mecalux 19
After reflecting on possible structural scenarios, a final scenario was built based on
feature iterative ramp-up and feature optimization
Analysis and
Deconstruction Baseline Scenarization Optimization
evaluation
20
Final solution builds on feature analysis and a carefully researched sorting criteria,
being built with an iterative ramp-up on previous best solution
PPs and POs ordering criteria Final solution rationale
¹Due to sorting by zone only sorting by 4 values, sorting by Weight as 2nd criteria makes this scenario only possibly fail thrice in weight validation/calibration
²Due to an average of 29 SKUs per picking pallet, grouping solution is only residually worse in picking time than sorting by demand
213
Final solution featured a 45% cost reduction from 195k€ to 105k€, with noticeable
results coming from ordering PO’s and PP’s and differentiated SKU treatment
Final solution results and feature impact on overall costs
195k € -20%1
-5%1
-5%1
-4%1
-6%1 105k €
-2%1
-3%1 ~0%1 -45%
Base Ordered Grouping Cs and Ds C+D Safety Bay 2 Pallets Full Final
Scenario PPs and as boxes Percentile Stock Optimized per Picker Pallets Scenario
POs Optimized
45% overall cost reduction 100% service-level achieved 0 pallets done off time
¹% as reductions to the base scenario
22
There was a focus on reducing warehouse and picking costs, which were the most
reduced (-71%) due to ordered picking strategy and warehouse dimension reduction
Picking Forklift Despicking Change in activity cost
Warehouse Transport
80 000,00 € Transportation 5%
51%
Main 44%
-
impact on
Change in
-35% -23% -13% -20% -24% -44% -14% -13% -5%
process1 Picking Warehouse Forklift Despicking Transportation
23
Storage organization results from an optimized storage length and efficient SKU
grouping with differentiated treatment
Grouping based on number of PO’s per SKU Storage functionality features
Increasing number of picking orders
S Flow through storages with ordered POs and PPs
minimizes picking costs
A B C D
2 passages allows for higher flexibility and more
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% escapeways
Pallets Boxes
I, L and U-shaped layouts were tested
Boxes
but didn’t show great solutions2
Picking
Position
Ground-level Ground-level 1st level
C+D
Storage
Layout
A
Maximum
Optimized SKU
storage Minimized costs
distribution1
occupation B
1Deep-dive next
2Analyzed in slide 18
24
Optimal SKU distribution in groups optimized cost savings and were based on
maximum storage occupation and iterative simulation testing
Rationale Total costs for varying Group C percentile
150 000,00 €
Group A percentile was based on top 20%1
Optimized +14%
SKU Group B and C percentile were substitute SKU’s, 145 000,00 €
distribution optimized:
+10%
• If Group B had more SKU’s, they would come 140 000,00 €
out of Group C
+7%
• Analysis conducted to check which percentile
Maximum 135 000,00 €
numbers would fit in a layout with maximum
storage
storage occupation2
occupation
• Tested results in AnyLogic simulation 130 000,00 €
25
Based on established number of SKU’s per group, a storage area optimization model
was built leveraging on operations research know-how
A+B can be in the same storages Based on number of bays and Arranged the warehouse in the
SKU’s per group least occupying area layout
If group D < group C, D will be in
the 1st level of C storages Calculation based on different
storage dimensions (A&B <> C&D)
10 000,00
24 bays was the optimal value for storage
area optimization, followed by 48 and 75,
Storage Area (m2)
8 000,00
with ~3150 m2 of storage area
6 000,00
After simulating in AnyLogic, it was clear that
4 000,00 24 bays was a better solution, due to having
2 passages, which created more flexibility
2 000,00 and escapeways for pickers
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79
Number of Bays
26
Drive-in pallet racking would be a possible implementation for the products of
Group C and D, as they have low rotation but should be further analyzed
Drive-in pallet racking
Warehouse costs reduced due to elimination of Higher rack damage rate due to the amount of
aisles, as the SKUs storage is compacted traffic in and out of the system
Unlimited storage depth Weak acessibility to pallets that are not in the front
position
¹Without considering forklifting operations, since D Group orders are unpredictable in a 15-day period and all the C products only require a picking operation
Stacker crane for pallets Trilateral stacker crane for pallets¹ Automated Pallet Shuttle
Automatic replenishment without the Adaptable to all pallet warehouses Rapid movement, which can optimize
need of an operator goods entrance and exit times
More freedom of movement due to
Reduction of possible errors during the 3-axis It can save staff costs, as the equipment
replenishment is automatic
¹To simplify, as this equipment has the same advantages as the Stacker crane for pallets, the pros and cons presented are just the additional ones.
28
The dynamic dispatch area was minimized by diminishing the time between issuing a
picking order and its shipping time (75’) while not compromising service-level
Dispatch area functionality features Sensitivity Analysis
Number of pallets done after time quantifies Number of pallets done after time
service-level, such that the lower, the better
As expected, increasing time between issuing picking order and
shipping time diminishes the number of pallets done after time
Optimal time of 75 minutes before shipping order
Also true at first for the number of pallets in dispatch due to the
~650 m2 is sufficient to fulfil current needs, allowing reduction of pallets that fail shipping and have to stay in dispatch,
for a 20% increase in daily number of picking pallets but will ramp up due to higher pallet time in stock
29
In order to test the final scenario, stress tests and overfitting tests were made to
ensure its quality when dealing with higher and varying demand
Analysis and
Deconstruction Baseline Scenarization Optimization
evaluation
30
Stress tests with double demand concluded that the model offers economies of scale
and full palleting is better in stress situations
-7%2
Expected Results
180 000,00 € Non Full Palleting Results
Full Palleting Results
160 000,00 €
80 000,00 €
¹Expected results based on double the average cost of both scenarios (in picking, forklift, despicking and transportation), maintaining warehouse costs
²Change in expected results to full palleting solution
³Despite this, current model supports a 20% increase in demand over the case’s picking orders
31
Overfitting tests were performed due to the reduced data amount, which allowed us
to conclude that the model maintained performance with test data
Based on case’s picking orders Base and final scenarios were Scenarios showed consistency in
tested several times with different relative results
Applied random variation to the data sets
number of boxes Model is not overfitted
32
A gradual implementation is required in order to identify possible risks and make a
smooth transition from the current warehouse to the new one
Dispatch Storaging Grouping
Implementation Plan
Analyze demand Apply storage layout and test Analyze grouping in-warehouse
historical data and for effectiveness and deviation effectiveness and configurate
resize dispatch area from expected results inbound strategy
Impact Likelihood of
Risk to Manage Mitigation Strategy
Assessment Occurring
33
Executive Summary
1. Sonae MC wants to redesign its PBS operation, looking for process redefinition and a layout suggestion
in order to reduce overall costs
2. Total costs can be minimized by decreasing picking, replenishment, warehousing, transportation and
despicking costs whilst keeping the same service level
3. In the base scenario, picking, warehousing and transportation account for 93% of total costs (195k€),
with picking, warehousing and service-level at sub-optimal values
4. Different features were tested individually on the base scenario to understand their impact on specific
and overall costs as well as assess possible synergies
5. Final solution builds on feature analysis and a carefully researched sorting criteria, being built with an
iterative ramp-up on previous best solution
6. Final solution featured a 45% cost reduction from 195k€ to 105k€, with noticeable results coming from
ordered picking strategy and warehouse dimension reduction
7. Storage organization results from an optimized storage length and efficient SKU grouping with
differentiated treatment and percentile optimization
8. Stress and overfitting tests concluded that the model offers economies of scale and maintains
performance with different data
34
Appendix 1: I-Shape, U-Shape and L-Shape Layouts
35
Minimizing costs
Redesigning processes
and layout
PBS Optimization
Professors: Pedro Amorim, Sérgio Castro
Thank you!
António Salgado | Luís Blanquet | Maria do Rosário Rocha
Maria João Durães | Ricardo Cabral
1M.EGI01