A Sensing and Prediction System for Quality- Projectcontext Quality Sensing (sensors, data loggers) Real-time Quality Controlled Flower Supply Chain Tracking & tracing (barcodes, RFID, wireless) Better quality for consumers Reduction of product waste Shorter lead times Internet Technology Better capacity utilisation 2 1
Background Internet of Computers Internet of Services (CLOUD) Internet of People (SOCIAL) Internet of Things (MOBILE) Emergence real-time condition monitoring barcode Passive RFID Active RFID and integrated sensors dataloggers WIFI, Bluetooth, 4G, GPS, LoraNet, etc. Wireless Sensor Networks 2
Towards smart autonomous objects I am product X at locatie L of Z Tracking & Tracing I don t want to stand besides that banana! I am thirsty! I am warm! Monitoring I am thirsty: water me within 1 hour! I am too warm: lower the temperature by 3 C Event Management My vaselife is optimal at a temperature of 4,3 C. Optimalisation I am too warm: I lower the cooling of my truck X by 2 C. Autonomy Source: Deloitte (2014), IT Trends en Innovatie Survey Scope of the pilot: up to optimisation! Prototype of an innovative system for real-time management of product quality in flower supply chains Based on expected vase life, flowering, ethylene damage, risks on Botrytis infection and other quality parameters Sensing and Prediction System for Quality Controlled Flower Supply Chains Tracking & Tracing Monitoring Event Management Optimisation Autonomy Source: Deloitte (2014), IT Trends en Innovatie Survey 3
Intended system functionalities Norms: - Logistics - Ambient conditions - Flower features - Expected quality Alerts norm deviations - Logistics - Ambient conditions - Flower features - Expected quality Identify deviations and generate Alerts Tracking & tracing items Item location Predicted Flower Quality Predicting quality trajectory Monitoring ambient conditions Current conditions Inspections current quality Product features Datamanagement Advices Advicing quality and logistics Project context and participants 8 4
Rough planning 2015 2016 Sep Okt Nov Dec Jan Feb Mar Apr Mei Jun Jul Aug Sep Okt Nov Dec 9 Forecasting flower quality by Bayesian Network Gitta ten Hoope Supply Chain consultant Royal FloraHolland 5
Principal Probabilistic graphical model based on Rev. Thomas Bayes (1702-1761) Sometimes the probability of a hypothesis is given before the event or evidence is observed Showed how to compute the probability of the hypothesis after some observations are made. 1. Rain influences Grass wet 2. Sprinkler is influenced by 3. In view of the fact that the grass is wet how big is the probablility that it rains? 07 July 2016 11 Self- learning and based on expert knowledge 07 July 2016 12 6
Advantages of Bayesian Networks Makes expert knowledge explicit and transparent Accepts qualitative and quantitative data Integration of Empiric models is possible Not all data are needed to be available always (Graceful degradation) Can be self- learning Feed back calculation (what did go wrong?) Calculations can be done directly Prediction of flower quality 1. Inventory of all influencers of flower quality 2. Answering the question of how big is the influence Packaging Origin Land/Region Cutting stems P(visible quality physical damages.) The probability of a good/ mediocre/ bad visible quality viewing the fact that physical damages are present/ absent. Physic. damages Visible Quality 7
Expert module 1. Different modules for each chain link 2. Modules can occur several times, depending on the real chain Grower Transport Handling Storage Diseases / pests 07 July 2016 15 How to built a Bayesian network? Sessions with experts: (Causality)diagram Filling the data tables with expert knowledge 8
Many thanks for your attention! GittaTenHoope@ royalfloraholland. com Cor.Verdouw@ wur.nl 17 9