AI and Its Applications in Manufacturing
Dr. Biplav Srivastava IBM Research – India Presentation to MEL 423 (Computers in Manufacturing Class) IIT Delhi, November 12, 2014
Outline l Artificial
Intelligence l Sample Applications l Manufacturing Domain and Planning l Improving Manufacturing with Data and Extracted Knowledge
Resources l AI – Summary of the sub-areas:
http://aitopics.org/topic/ai-overview – Russell and Norvig, AI – A Modern Approach – http://en.wikipedia.org/wiki/ Artificial_intelligence
Artificial Intelligence (AI) l Intelligent
Agents - Build useful systems
– Symbolic: logic, reasoning, search – Statistical: machine learning, Bayes rules
l Cognition
– Understand working of brain
Sample Systems l ASIMO
Copies Dance Moves (video) l Chat with Ramona (link) l IBM Watson (video)
Sample AI Applications · · · · · · · · · · · · · · · · · · · · · · ·
Agriculture & Natural Resources Archaeology Architecture & Design Art Artificial Noses Assistive Technologies Astronomy & Space Exploration Automatic Programming Automotive Industry Autonomous Vehicles Aviation Banking & Finance Bioinformatics Biometrics Business & Manufacturing Chatbots Chemistry Decision Support Systems Earth & Atmospheric Science Engineering Design Fraud Detection Hazards & Disasters Knowledge Management
· · · · · · · · · · · · · · · · · · ·
Law Law Enforcement & Public Safety Machine Storytelling Marketing Medicine Military Music Networks Oil & Gas Politics & Foreign Relations Recommender Systems Robots in the Home Robots in the Workplace Science & Mathematics Smart Houses & Appliances Social Science Software Engineering Spam Filtering Surveillance
· Telecommunications · Transportation & Shipping
Software System Referenced Data
Move More Business Logic To Declarative Data (policy)
Business Logic Processing
Reading Input
Produce Output
Environment Input 7
Output © 2014 IBM Corporation
Example: Taking Care of a Baby Individual’s Extension
Agent Expected behavior: •
• 8
Inform • • • Do • • •
Alert when crying Alert when awake Alert when idle Raise temperature of room Play music …
© 2014 IBM Corporation
Example: Taking Care of a Senior Assisted Cognition
Agent Expected behavior: •
• 9
Inform • • • Do • •
Alert when idle Alert when away from known locations Alert when checkup/ medicines due Send body parameters periodically … © 2014 IBM Corporation
Example: Taking Care of Oneself Personal Digital Assistants Agent
Expected behavior: •
•
10
Inform • When missing meetings • When missing social commitments • Reminding of priorities • … Do • Make all cancellations / re-bookings when schedule changes • Find alternatives to current decisions and give choices (e.g., traffic) • …
© 2014 IBM Corporation
What Are Intelligent Agents? Agents are active, persistent software components that perceive, reason, act, and communicate. (Huhns and Singh) Software that assists people and acts on their behalf
Agents can help people and processes Agents are used for automation and control finding and filtering information personalizing your environment negotiating for services automating tedious tasks taking actions you delegate learning about you over time collaborating with other agents capturing individual and organizational knowledge sharing knowledge
01/23/12
finding and fixing problems automating complex procedures finding "best fit" procedures pattern recognition and classification predictions and recommendations negotiate and cooperate with other organizations' agents
11
IBM ABLE Agents Overview
Planning & (Classical Planning) (Static) Environment (Observable)
Goals perception (perfect)
action (deterministic) What action next?
I = initial state [I]
G = goal state Oi
Oj
(prec) Ok
Slide courtesy: Prof. Subbarao Kambhampati, ASU
Om
Oi
(effects)
[G] Dr. Biplav Srivastava
Simple Planning Example Blocks World Robot arm
A A
B
Initial State
B
Blocks
Goal State
All robots are equivalent
Representation A
B
States: ((On-Table A) (On-Table B) …) Actions: ((Name: (Pickup ?block ?robot) Precondition: ((Clear ?block) (Arm-Empty ?robot) (On-Table ?block)) Add: ((Holding ?block ?robot)) Delete: ((Clear ?block) (Arm-Empty ?robot)))…)
A
B
Planning Process Pick-up B R1 Clear A Clear A
Pick-up B R2
Clear B
Pick-up A R1
Clear B
On-Table A
On-Table B
On-Table B
Arm-Empty R1 Arm-Empty R2
Put-down A R2
Holding A R1
On-Table A
Pick-up A R2
Put-down A R1
Stack A B R1 Stack A B R2
Holding A R2 Arm-Empty R1 Arm-Empty R2
Initial State Level P-0
On A B
Level A-0
Level P-1
Level A-1
Goal State Level P-2
Manufacturing Illustration
Slide courtesy Shart Sood’s slideshare presentation on Production and Operations Management, At: http://www.slideshare.net/technomgtsood/production-operations-management
Planning in a Manufacturing Process
IMACS, A System for Computer-Aided Manufacturability Analysis Slide by Marc Berhault, 2003
Complexity Example
IMACS, A System for Computer-Aided Manufacturability Analysis Slide by Marc Berhault, 2003
IMACS Approach Illustration
IMACS, A System for Computer-Aided Manufacturability Analysis Details at: http://www.cs.umd.edu/projects/cim/imacs/
Destination Control of Elevators l
Schindler Lifts. – Central panel to enter
your elevator request. – Your request is scheduled and an elevator is assigned to you. – You can’t travel with someone going to a secure floor, emergency travel has priority, etc. l
Modeled as a planning problem and fielded in one of Schindler’s high end elevators.
Reference: https://user.enterpriselab.ch/~takoehle/ publications/elev/elev.html
Improving Manufacturing with Knowledge (Beyond Robotics) l Enterprise
Resource Planning l Enterprise Integration l Open Data and Analytics
(Manufacturing) Key Information Flow
Slide courtesy Shart Sood’s slideshare presentation on Production and Operations Management, At: http://www.slideshare.net/technomgtsood/production-operations-management
Analogical Example: Documenta3on in Health Care
Doctor’s Office Medical History
Prescription
Medical Claims, Prescription
Patient Bills, Policy Updates
Pharmacy Drug Claims, Prescription
Insurance Problem: Hand-offs between role-players are inefficient and failure-prone
Enterprise Resource Planning: Packaged Applica3on l
Packaged Applica3on – Off-‐the-‐shelf soBware to manage common business func3ons like accoun3ng,
payment and receivables, order management, customer management; or industry-‐ specific func3ons like clinical trial (pharmaceu3cal), drilling (mining, oil & gas) – Businesses buy these soBware and then engage service providers to tailor them – Enterprise Resource Planning (ERP) is a specific class of packaged applica3ons l
Market size (according to AMR Research [AMR 2008]) – The total market size for ERP soBware is currently $34.4B. SAP leads with 42 %,
followed by Oracle (23%), The Sage Group (7%), MicrosoB Dynamics (4%), and others. – Spending on services including consul3ng, integra3on and support for Oracle, SAP, and other business applica3on vendors, called packaged enterprise applica3on services, was $103B for 2007, and expected to reach $174B by 2012. – IBM is a prominent service provider for SAP and Oracle. l
Trend is to move from Legacy applica3ons to Packaged Applica3ons
Packaged Applica3ons:
Pre-‐built, Configurable, Business Processes Automa3on SoBware Packaged Application
SAP
Oracle
Cross-industry: ERP, CRM, SCM, PLM, SRM, HCM, eProcurement
√
√
Industry specific solutions
> 50: Aerospace &Defense, Automotive
> 50: Aerospace &Defense, Automotive
(3), Chemicals, Consumer Products (5), Construction, High Tech (4), Industrial, Life Sciences, Mill Products (5), Mining, Oil & Gas, Media (4), Services (2), Telco, Utilities (5), Waste & Recycling, Travel & Logistics (4), Banking, Insurance, Public Sector (2), Defense (2), Healthcare, Education & Research (2), Retail, Wholesale Distribution
(2), Chemicals, Consumer Products, Construction, High Tech (6), Industrial (3), Life Sciences, Mill Products (5), Mining, Oil & Gas (3), Media (4), Services (4), Telco, Utilities (4), Waste & Recycling, Travel & Logistics (5), Banking, Insurance, Public Sector (3), Defense (2), Healthcare (2), Education & Research (3), Retail, Wholesale Distribution
Packaged Application SW Revenue: $38B (2008)
Source: AMR Research 2008
Market Share Analysis: ERP Software Worldwide, 2012 authored by Chris Pang, Yanna Dharmasthira, Chad Eschinger, Koji Motoyoshi and Kenneth F. Brant.
Health with Data.gov.in Data (Since Feb 2013)
Conjectures: - DS1 and DS2: - Is gap in staff (e.g., pharmacists) correlated with gap in infrastructure (e.g., primary health centers)? - Similarly other variants on staff and infrastructure Manufacturing Implication – manufacture vaccines that can taken without staff - DS3 and DS4: - Is increase in booster dose correlated with children mortality? That is, do states having more women covered also have lower mortality? Manufacturing Implication – where to market booster doses - DS1 and DS3: - Is gap in staff (e.g., pharmacists) correlated with number of women getting booster shots? - DS2 and DS4: - Is gap in infrastructure (e.g., health centers) correlated with children mortality? 26
Data sets * DS1: http://data.gov.in/dataset/ pharmacists-laboratorytechnicians-and-nursing-staffprimary-health-centrescommunity-health * DS2: http://data.gov.in/dataset/ shortfall-healthinfrastructure-2011population-provisionalindiaas-march-2011 * DS3: http://data.gov.in/dataset/ number-women-giventt2booster * DS4: http://data.gov.in/dataset/ under-5-mortality-rate-u5mr
Putting All Together for Manufacturing Use AI for deciding l l l l
What should one manufacture? For whom, at what price point and with what process? How should the product be maintained? How does one reduce wastage? How can we bundle services with it and increase service quality?
Relevant Key AI Areas Knowledge Representation l Ontology l Reasoning and Logic l Machine Learning l Optimization, Scheduling l Planning l Mechanism Design, Auction l