DSTA-MIT ESD Armament Safety Project 22 January 2007

DSTA-MIT/ESD: Armament Safety Project 22 January 2007 • MIT Team – Prof. Stuart Madnick (smadnick@mit. edu) – Dr. Michael Siegel – Lynn Wu – Nathan Minami (MIT & US Army) – Allen Moulton – Mihai Lupu (SMA) – Assistance from: • Prof. Aykut Firat (MIT & Northeastern University) • Prof. Hongwei Zhu (MIT & Old Dominion University) © MIT, 2007 1

DSTA-MIT/ESD: Armament Safety Raw Knowledge High-Level Project Overview Sources Knowledge Repository SEM 2007 -01 -13 High Level Policy choices and Interventions Safety Indicators External sources Safety Forecasting Model (2) Internal sources: Assessment Reports, live firing trials data, incident investigation reports, acceptance test data, qualification trials reports, hazard database for various items, WDA, defect rates, range & training schedule, weapon & ammo procurement and ages, etc. Knowledge Re-Use Data Extraction & Semantic Reconciliation Technologies (1) Detailed Indicators Funding = Training Schedule Pressure Ammunition Age 2

Knowledge Management Activities [Phase 1 and 2] Diverse Raw Knowledge Sources Knowledge Repository (a) Raw Knowledge Retrieval (e. g. , specific piece of knowledge) [Phase 1] External sources (b) Processed Knowledge Retrieval (e. g. , comparisons and/or aggregations of knowledge) [Phase 1] Knowledge Storage Internal sources Knowledge Capture [Phase 1] (c) Projections based on Knowledge (e. g. , Forecast and/or Consequences) [Phase 2] Knowledge Re-use 3

Example of Basic Knowledge Capture and Re-Use • Goal: Demonstrate use of MIT extraction and analysis technologies • Constraint: Difficult to get access to diverse data related to Singapore Armament Safety (so far) • Approach: Demonstrate using public sources and interesting strategy questions • Example issue: What is Singapore military expenditure per capita? • How do others compare (+ or - %) with Singapore: – – Israel = USA = UK = Malaysia = 4

Good News – Data Publicly Available (https: //www. cia. gov/cia/publications/factbook/geos/sn. html) Population Economy Military 5

But it is scattered … must be located … • … and not in exactly easy to use form … • Must do calculations: – Military expenditures = 4. 9% (of GDP) x $126. 5 billion = $6. 2 billion – per capita = $6. 2 billion / 4, 492, 150 people = $1379. 85 per capita • Need to be careful about scale factors – (one of our team members was off by a thousand) • Amount is in US$ … what if you want in SG$? 6

Easier way: Cameleon Web Extraction Tool ( http: //coin. mit. edu/DSTA/Mihai/cameleon. html Results as tsv (data record) coinsmadnick 6671 coin ) Results as XML file - - Singapore 4, 492, 150 126. 5 4. 9% 7

But also want: (1) multiple countries and (2) military per capita calculation • First: Multiple countries. – Combine Cameleon web extraction capability with IBM DB 2 Federated Database Server • Technical details explained later SQL Query: Select country, population, GDP, Mil. Expend. Percent from cia where country in ("Singapore" , “Israel”, “United States” , “United Kingdom” , “Malaysia”) 8

Now need military per capita calculation • DB 2 can provide the computation based on the data being extracted by Cameleon Select country, Mil. Expend. Percent * GDP * 100000 / population from cia where country in (Singapore" , “Israel” , “United States”, “United Kingdom” , “Malaysia”) Note: Is USA really only $1. 67 per capita and Singapore $1379 per capita? 9

Need to adjust for GDP “billion” and “trillion” • Use an auxiliary table – scalefactor text scale Billion 100000 Trillion 1000000 • Now can use database “Join” Select country, Mil. Expend. Percent * GDP * scalefactor. scale / population from cia, scalefactor where country in ('Singapore', 'Israel', ' United States', 'United Kingdom', 'Malaysia') and scalefactor. text=cia. gdp_unit 10

That is great: but want to graph comparisons • Combine Cameleon web extraction capability with Excel spreadsheet graphing capabilities • Can put SQL queries to Cameleon in Excel Query automatically assembled. . And sent to Cameleon 11

Excel provides nice graphics • Excel can display results in both table form and graphics 12

Need to adjust scale factors • Some GDP’s are stated in billion and some on trillion – by the way: using USA definition of “billion” • Can use conditional definition for military_expenditure per capita that handles scale factor adjustment. 13

Final results of “military expenditure per capita” – with scale factor adjusted • Using Excel formula with conditional definition (Conditional definition) 14

Much easier to “visualize” graphically • All kinds of additional calculations and graphics possible 15

Military expenditure in multiple currencies • The data in CIA Fact book is all in US dollars. – Might be more helpful to view in other currencies. • Currency conversion web sites also exist: www. oanda. com – We “wrapped” it – called it “hzoanda” Select any currency Query automatically assembled. . And sent to Cameleon 16

Military expenditure in choice of currencies 17

These solutions can be even easier through COIN context mediation • MIT has developed a Context INterchange (COIN) technology that can automatically: – Adjust scale factors and – Provide answers in SG$ (or any currency), etc. • This query: Select country, Mil. Expend. Percent * GDP * scalefactor. scale / population from cia, scalefactor where country in ('Singapore', 'Israel', ' United States', 'United Kingdom', 'Malaysia') and scalefactor. text=cia. gdp_unit [currency conversion not shown] • Could be reduced to: Select country, Mil. Expend. Percent * GDP / population from cia where country in ('Singapore', 'Israel', ' United States', 'United Kingdom', 'Malaysia') • This part of project cannot be demonstrated today due to recent hardware problems that disrupted system and corrupted some software. 18

Information may be scattered across multiple sources • What if want military expenditure per soldier? • Can get size of armed forces from http: //en. wikipedia. org/wiki/List_of_co untries_by_size_of_armed_forces • Need to combine this data with the information from CIA Fact Book to get military expenditure per soldier 19

Can do “Joins” across multiple sources select cia. country, GDP_unit, armed_forces, Mil. Expend. Percent * GDP* scalefactor. unit / armed_forces/1000 as milpersoldier from cia, armforces, scalefactor where cia. country in ('Singapore', 'Israel', 'United States', 'United Kingdom') and armforces. country = cia. country and scalefactor. text=cia. gdp_unit 20

What about Armament Knowledge? • More good news … • There is (we believe) lots of armament information on SAF intranets – Though we have not been able to access much yet • There is lots on public Internet – for fee – Jane’s Information Group (http: //www. janes. com ) • There is lots on public Internet – for free – Small arms, Tanks, Artillery, etc • We would need assistance of Subject Matter Experts to determine needed information • In meantime, here is a brief demonstration … 21

Some Example Questions? • Firearms & RPG: How do the weights of an AK-47, AK-74, M-15 A 4, and M 16 A 2 compare ? • Tanks: Where is the Centurion Mk 13, M 1 A 2, and T-54 made and what is the range and speed of each? 22

How do the weights of an AK-47, AK-74, M-15 A 4, and M 16 A 2 compare ? • Information scattered across multiple sites: http: //www. warfare. ru/? lang=&catid=246&linkid=1814 http: //www. armalite. com/sales/catalog/rifles. htm http: //www. fnmfg. com/products/ http: //www. armalite. com/sales/catalog/rifles/m 15 a 2. htm http: //www. fnmfg. com/products/m 16 fam/m 16 a 2. htm 23

Can be accomplished by a single query select weapon, weight_unit from armlite where weapon in('AK-47', 'AK-74', 'M-15 A 2', 'M 16 A 2') union select weapon, weight_unit from warfareru where weapon in('AK-47', 'AK-74', 'M-15 A 2', 'M 16 A 2') union select weapon, weight_unit from fnmanufacturing where weapon in('AK-47', 'AK-74', 'M-15 A 2', 'M 16 A 2') * * This query could be further simplified by using a database View of the 3 db’s. 24

Where is Centurion Mk 13, M 1 A 2, and T-54 made and what is the range and speed of each? • Even when all the information is on a single site, it is hard to pull together for comparisons http: //www. fas. org/man/dod-101/sys/land/row/tanks. htm 25

But can be done with a simple query select weapon, range, speed from tanks where weapon in ('CENTURION Mk 5', 'T-55', 'M 1 A 2', 'M 1') 26

How does it work? Web-Wrapper Techno User or Program (via SQL Query) Select Edgar. Net_income From Edgar Where Edgar. Ticker=intc and Edgar. Form=10 -Q Web page spec file * SQL Side Ticker INTC Web Wrapper Generator HTML Side Net Income 1, 983 Data record returned * Spec file contains: Schema, Navigation rule 27 and Extraction rules.

Recent Technology Extensions to Support DSTA Effort • Cameleon Studio – Makes it much easier/faster to create spec files • Incorporate Cameleon into IBM’s DB 2 Federated Database System * – To handle much more complex SQL queries • Calculations • OR and IN features – Required creation of a general-purpose “capability handling” engine • Web sites are more limited than databases (e. g. , may only take one “key” at a time, only certain fields can be used as a “key”, etc. * For more information see: Lynn Wu, Aykut Firat, Stuart Madnick, Tarik Alatovic, “Querying Web-Sources within a Data Federation”, Proceedings of the International Conference on 28 Information Systems (ICIS), Milwaukee, Minnesota, December 2006.

Cameleon Studio tool makes spec file creation (relatively) easy 29

Previous Cameleon + POE Architecture • Our Planner, Optimizer, Executioner (POE) facility extended Cameleon web wrapper POE handles complex SQL Data Application Output in Retrieval HTTPClient Relational Front End -Divides query into simpler single source queries Planner, Optimizer, -Dispatches the queries to core Executioner (POE) Cameleon or SQL db Simple SQL Query - Assembles results together Desired Format Core Cameleon Extraction Regular Expression Engine & Output Format Query Handling Spec File Parsing Authentication, etc Web or Database multi-source query (including “join”s): Spec Files Registry Problem - POE does not handle full SQL needed: - No calculations ( e. g. , gdp/capita) - Allows JOINS and UNIONS but not IN or OR 30

IBM’s DB 2 Federated Database System • Uses wrappers to access non-relational data sources. • DB 2 first decomposes the original query into query fragments and then sends them to wrappers. – Some generic wrappers provided (e. g. , for XML sources) – For arbitrary sources (such as web sites), custom wrappers must be created • Wrapper sends the result back to DB 2 which then assembles the final results. DB 2 XML Wrapper (Adapted from IBM). 31

IBM’s DB 2 Federated database system architecture Query: Select. . from s 1, s 2, s 3 … Federation Engine IBM DB 2 Wrapper: Request-Reply Protocol Wrapper For each web site (S 1, S 2, S 3), a wrapper must be custom crafted. Wrapper for S 1 Wrapper for S 2 Wrapper for S 3 Capability Handler Data Extraction Web Sources S 1 -website S 2 -website S 3 -website 32

Our Solution: (1) Cameleon as general-purpose wrapper for web sources (2) with new capability engine Two-Layered Architecture Three-Layered Architecture - current IBM solution with general-purpose engines Query: Select. . from s 1, s 2, s 3 … Federation Engine IBM DB 2 Wrapper: Request. Reply Protocol - Wrapper for S 1 Capability Handler Wrapper: Request -Reply Protocol Wrapper, Capability Engine Wrapper for S 2 Capability Handler Wrapper for S 3 Data Extraction Engine Capability Record Declaration CR for S 2 CR for S 3 DE for S 1 Data Extraction Spec Files DE for S 2 DE for S 3 Web Sources S 1 -website CR for S 1 Capability Handler Cameleon Data Extraction Query planning with capability declaration S 2 -website S 3 -website S 1 -website S 2 -website S 3 -website 33

Phase 2 – Evaluation of Policy/Intervention Options for Improving Armament Safety • Approach: Use of System Dynamics (SD) Modeling – Used by US Government/Military for many purposes – Recent example – evaluating interventions on “State Stability” (see http: //web. mit. edu/smadnick/www/wp/2005 -13. pdf ) • Goal: Develop a system dynamics model for Singapore Armament Safety – Represent causalities and linkages – Uncover root causes • Outcome: Model can be used to predict the likelihood of future accidents and help evaluate policy options and preventive measures. 34

Philosophy of System Dynamics • Every action has consequences • Often through complex feedback loops Do you feel crowded in – and frustrated? 35

See if you can get a bit more space by pushing on that wall … 36

Oops … 37

Quick Primer: The What (and Why) of System Dynamics Consider the domain of Software Development • Classic “knee jerk” reaction to a project behind schedule is to add people. • “Brooks Law” noted that “Adding people to a late project, just makes it later” (because the new people must be trained, this takes productive people off the project). • Both of these points are fairly well-known by most software developers – but both are much too naïve. There are many other factors: length of project, type of project, expertise of staff available, approach to and time needed to do training, stage of project, etc. Over the years, all of these individual factors have been well-studied individually – but how do they interact ? • System dynamics makes it possible to model & study the dynamics of these interdependencies. Many non-obvious outcomes have been found (e. g. , sometimes Brooks is wrong! Important to know when and Why? ) Source: Software Project Dynamics: An Integrated Approach, by T. K. Abdel-Hamid and S. Madnick, Prentice-Hall, 1991. 38

History of System Dynamics SDM used as modeling & simulation method over 30 years • Designed to eliminate limitations of linear logics and oversimplicity • Typical human assumptions and behaviours • Based on understanding system structure, behavior patterns, interconnections of positive & negative feedback loops, and intended & unintended consequences of action SDM has been applied to numerous domains, e. g. , • • • Software development projects Process Improvement projects Crisis and threat in the world oil market Stability and instability of countries … many others … SDM helps to uncover ‘hidden’ dynamics in system • • • Helps understand ‘unfolding’ of situations, Helps anticipate & predict new modes Explore range of unintended consequences 39

Armament Safety Example: consequences and proceeding actions Held up due to traffic accident Was busy dealing with an earlier accident Safety officer arrived late Hypothetical Example Training started late Training was behind schedule Did not stop to cool off Weapon had overheated Weapon had jammed, which damaged ammunition Damaged ammunition Chamber explosion Soldier injured 40

Preliminary Example of Armament Safety SD Causal Model Developed based upon Singapore reports provided and Subject Matter Expertise (SME) provided by US Army Major Nathan Minami, Company Commander 25 th Infantry Division and West Point instructor. 41

Plans for Phase 2 • Focus: Armament Safety Projection & Mitigation (e. g. , Understanding Consequences) - Maybe include Vehicle Safety • Needs: – Access to Subject Matter Experts (Plan to also explore US Army sources) • To develop causal model • To determine “tipping points” and other “non-obvious” behaviour – Access to Data • To parameterize model for simulation 42