1. Shelf life prediction by intelligent RFID -
Dynamics in Logistics
Shelf life prediction by intelligent RFID -
Technical limits of model accuracy
Jean-Pierre Emond, Ph.D.
Associate Professor,
Co-Director
UF/IFAS Center for Food Distribution and Retailing
University of Florida
Reiner Jedermann Walter Lang
IMSAS Institute for Microsensors, -actuators and systems
MCB Microsystems Center Bremen
SFB 637 Autonomous Logistic Processes
University of Bremen

2. Outline CFDR / University of Florida IMSAS / University Bremen
Evaluation of quality
Case Study “Strawberries”
IMSAS / University Bremen
Integration of quality models into embedded hardware
Intelligent RFID
Feasibility / required hardware resources

3. Center for Food Distribution and Retailing

4. Laboratory evaluation of shelf life models
Several attributes have to be tested
color
firmness
aroma / taste
vitamin C content
(Nunes, 2003)

5. Strawberries – Case Study
Joint project between Ingersoll- Rand Climate Control and UF
Temperature sensors were placed inside and outside the load at all locations in the trailers
Quality was assessed from beginning to end
How retailers evaluate the quality of a shipment?
Economic impact of monitoring temperature and quality prediction

6. Strawberries – Case Study
= 3 full days
= 2 full days
= 1 full day
= 0 day
RFID Temperature Tag + Prediction Models

7. Strawberries – Case Study
FEFO = First expires first out
= 3 full days
= 2 full days
RFID + Models decision:
2 pallets never left origin
2 pallets rejected at arrival
5 pallets sent immediately for stores
8 pallets sent to nearby stores
7 pallets with no special instructions (remote stores)
= 1 full day
= 0 day
RFID Temperature Tag + Prediction Models

8. Strawberries – Case Study
Days left
Number of pallets
Waste random retail
Waste (RFID + Model)
(Recommendation) 2
91.7%
(rejected)
(don’t transport) 1 5
53 %
(25%)
(sell immediately) 8
36.7%
(13.3%)
(nearby stores) 3 7
10%
(10%)
(remote stores)
Results at the store level (22 pallets sent)

9. Revenue and Profit Strawberries – Case Study Actual RFID + Model
COST $49, $45,480
PROFIT ($2,303) $13,076
Revenue and Profit

10. The idea of intelligent RFID
Avoid communication bottleneck by pre-processing temperature data inside RFID
Temperature curve
Function to access effects of temperature onto quality
Only state flag transmitted at read out

11. Chain supervision by intelligent RFID
Step 1: Configuration
Step 2: Transport
Step 3: Arrival
Step 4: Post control
Handheld Reader
Manufacturer
Reader gate
Measures and stores temperature
Calculates shelf life
Sets flag on low quality
List
Temperature
Shelf life
Transport Info
Full protocol

12. Modeling Approaches Different model types
Reaction kinetic model (Arrhenius)
Different model types
Tables for different temperatures
Differential equation for bio-chemical processes
d[P] / dt = −kPPO*[P]
d[PPO] / dt = kPPO[P] − kbrown*[PPO]
d[Ch] / dt = kbrown*[PPO]

13. Example Table Shift Approach
Only curves for constant temperature are known
How to calculate reaction towards dynamic temperature?
Interpolate over temperature and current quality to get speed of parameter change
Temperature Change from 12 °C to 4 °C

14. Model accuracy Measurement tolerances
Parameters like firmness or taste have high measurement tolerances
Question: Is this table shift approach allowed?
Yes, if all entailed chemical processes have the similar activation energies (similar dependency to temperature)
Otherwise testing for the specific product required

15. Simulation
Comparison of reference model (Mushroom DGL) with table shift approach
Parameter tolerances 1 % and 5%

16. Hardware Platforms Wireless sensor nodes Goal Tmode Sky from Moteiv
Own development (ITEM)
Goal
Integration into RFID-Tag
Comparable to RFID data loggers

17. Required Hardware Resources
Type of Resource
Calculation of Arrhenius equations
Look up table for Arrhenius model
Table-Shift Approach
Processing time
1.02 ms
0.14 ms
1.2 ms
Program memory
868 bytes
408 bytes
1098 bytes
RAM memory
58 bytes
122 bytes
428 bytes
Energy
6 µJoule
0.8 µJoule
7 µJoule

18. Power consumption per month Typical battery capacities
Available Energy
Power consumption of model is not the issue
Multi parameter models are feasible on low power microcontroller
Reduce stand by current
Power consumption per month
Update every 15 minutes
(Table shift / 1 Parameter)
20 mJ / month
Stand by current of MSP430
(1µA at 2.2V)
5700 mJ / month
Typical battery capacities
Button cell
300 … 3000 J
Turbo Tag (Zink oxide battery)
80 J

19. Summary and Outlook
Case study (strawberries) showed the potential to reduce waste and increase profits
Quality evaluation of the level of RFID tags is feasible
Testing on existing hardware of sensor nodes
Development of new UHF hardware required

20. Thanks for your attention
The End
Thanks for your attention