© Her Majesty the Queen in Right of Canada (Department of National Defence), 2019

Dr. Peter Young CORA / CFMWC ORT

Presentation to the 36th ISMOR 23 – 26 July 2019

Leveraging physics-based simulations for Operational Analysis: Task Group Air Defence Case Study

Background: OA tasking to study Task Group Air Defence

Late 2017 the Canadian Forces Maritime Warfare Centre (CFMWC) received an Operational Analysis (OA) tasking to study naval Task Group (TG) Air Defence (AD)

CFMWC stakeholders:

Above Water Battlespace (AWB): expertise in Anti-Air Warfare (AAW) weapon systems, tactics and doctrine

Modelling and Simulation (M&S): provision of computing infrastructure and M&S expertise supporting the CFMWC battlespaces

Operational Research Team (ORT): 3 defence scientists supporting AWB, Underwater Battlespace (UWB) and Joint Technology and Innovation Battlespace (JITB)

This presentations provides an overview of how physics-based simulations employed by M&S to support AWB are being used to underpin the TG AD OA study

1

Acknowledgements to the officers and staff of CFMWC AWB, M&S and ORT for their contributions through the TG AD study

Presentation Overview

Overview of CFMWC modelling and simulation capability

Modelling capability

Computing infrastructure

Case Study: Task Group Air Defence

Study design considerations

Analysis approach

MEZ construction using physics-based simulations

Ship stationing analysis with illustrative results

Summary with observations

Concluding remarks

2

All data and examples shown are for illustrative purposes only and do not reflect actual systems.

CFMWC modelling and simulation capability

3

fidelity

low

high

abstraction

high

low

tactical

system

subsystem

operational

strategic

one vs. one

few vs. few

one vs. few

many vs. many

force on force

propagation

radar ballistics sonar

Underwater Warfare

Campaign

Joint

Task Group

physics-based Monte Carlo simulation

acoustics fly-out 3 DOF 6 DOF

{

Gunnery

Air Defence

platforms/weapons sensors

C2

atmosphere

EM

CFMWC computing infrastructure

4

syndicate rooms briefing/training theatre client

workstations

servers

computing cluster

Network Configuration

databases

gateway

fidelity

low

high

abstraction

high

low

tactical

system

subsystem

operational

strategic

one vs. one

few vs. few

one vs. few

many vs. many

force on force

propagation

radarballistics sonar

UnderwaterWarfare

Campaign

Joint

Task Group

physics-basedMonte Carlosimulation

acousticsflyout3 DOF6 DOF

{Gunnery

Air Defence

platforms/weapons sensors

C2

atmosphere

EM

Industrialised M&S supporting the Royal Canadian Navy

5

Applications: Concept Development & Experimentation (CD&E)

Training

Tactics development

Doctrine development

Operational Test & Evaluation (OT&E)

Planning

Post-trial reconstruction and analysis

Requirements development & verification

Operational Analysis

syndicate roomsbriefing/training theatre

servers

computing cluster

Network Configuration

databases

gateway

storage processing

servers

High Throughput Computing (HTC) for M&S

6

1. model & data deployment

3. data parsing & storage

2. model execution

4. result queries

1000+ processors

HTC enables Monte Carlo simulation as a viable means for exploring large parameter spaces using high fidelity models

2+ Petabytes

Parameter space dimensionality, run-time speeds and data management still provide limitations

Case Study: Naval Task Group Air Defence

7

Air threat • Fighter Bomber • Anti-Ship Missile (ASM)

Air Defences • Long Range (LR) Surface-to Air Missile (SAM) • Short Range (SR) SAM • Naval Gun

High Value Unit (HVU)

Air Defence (AD) Ships

Study design considerations

8

Threat presentation • Threat type and number • Main threat axis • Spacing and timing about threat axis

Task Group • Numbers and types of ships • AD ship capabilities • C3I systems • Mission • Threat assessment • Firing policies • Ship stationing Atmospheric conditions

• Temperature/Pressure • Ducting • Wind • Water surface

Use physics-based simulations to perform engagement assessments

Use high level “OA” approach to analyse TG configurations

AD Ship • Search radar • Tracking radar • SAM • Seeker • Propulsion • Flight dynamics •Warhead

Threat • Seeker • Flight dynamics • Signature • Vulnerability

System /subsystem aspects Scenario / study aspects

fidelity high

abstraction high

Key metric: SAM expenditure to counter the threat raid (expressed as cost)

Problem parameter space

9

Number of runs: NR = NC x NMC ≈ 45x109

11x21

Grid box for ship: 21x21 21x21

Number of TG configurations: NTG = (11 x 21) x (21 x 21) x (21 x 21) ≈ 45x10

6

Number of threat axes: NTA = 20

Number of threats: NT = 4

Number of ships: NS = 3

Excessive, even with HTC Number of engagements: NE = NT x NR ≈ 18x10

10

AD ship HVU threat

Number of cases: NC = NTG x NTA ≈ 9x108

Number of firing policies: NFP = ?

PK = 0.7

PS = 0.3 Monte Carlo runs per case: NMC = 50

AD Ship configuration: radar, C3, LR SAM, SR SAM Threat: Sub/supersonic, manoeuvring/non-man, seeker

Analysis approach

10

2.a Employ MEZs in a lookup manner to determine optimal firing policies

Coordinated AD: shoot1 – look –shoot2

XX

ship1 ship2

2.b Conduct ship stationing analysis 3. Cross-validate results with and perform focused runs using high fidelity simulations

model execution

cross-validation

focused runs

1. Use high fidelity simulations to construct Missile Engagement Zones (MEZs)

X

results

HVU

MEZ

Air Defence Model 6 DOF Missile Model

Air Defence Model 6 DOF Missile Model

Commercial Monte Carlo simulation with integrated high fidelity models from DRDC and Allied partners

X

Step 1. MEZ construction using physics-based simulation

11

S1 initiate firing

S1 launch

S1 intercept

S1 assessment

timeline threat

velocity v

time to initialise

fly-out time

time for assessment

Descretised MEZ

cross-range

down- range

Grid box for MEZ: 41x61

Number of grid pts: NG = 21 x 61 = 1,281

model execution

LR & SR MEZs • 1 to 2 days for model execution • 1 week turn-around, including data

parsing and post-analysis

Monte Carlo runs per case: NMC = 50

Number of runs per threat type: N = NG x NSAM x NMC = 128,100 *

Number of SAM (LR & SR): NSAM = 2

Air Defence Model 6 DOF Missile Model

HVU

XX

* Plus other factors: e.g. threat flight profile and ducting height

Step 2. Ship stationing analysis

12

Inputs • SAM – threat MEZs •Number of AD ships •Number of threats • Range of TG configurations • Range of threat axes • Threat axis assumptions • Firing doctrine constraints • AD coordination

Functionality • 2.a For each threat axis/TG configuration,

determine optimal firing policy to achieve a desired PK against each threat whilst minimising SAM expenditure • 2.b For each threat axis assumption,

determine the TG configuration that minimises mean missile expenditure

Outputs •Optimal firing policy for each

threat axis/TG configuration •Optimal TG configuration for each

threat axis assumption

Ship Stationing Model

MEZs TG threats

0

0.05

0.1

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

0

0.2

0.4

0.6

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

threat axis assumptions

0

0.05

0.1

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

0

0.2

0.4

0.6

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

aggregation

1

2 3 1

2 3 4

Cmn = 5.8 Cmn = 6.2

FP: T1: S1S1 T2: S1S1 T3: S2LS1 T4: S2LS1 C = 6.6 PK = 0.9

2.a 2.b

X

SSM showing illustrative MEZ for a crossing target

13

MEZ captures key aspects of a complete single shooter – single target engagement • Detection by a search radar • Tracking by a fire control radar • Launch and fly-out of the SAM

with associated fly-out time • Intercept of the SAM with the

target with resultant single-shot PK

Launch points Intercept points

MEZ dependencies • Target type and flight profile

(height & velocity) • Search and tracking radar

performance • SAM guidance and flight

capabilities • SAM warhead effectiveness against

the target • Environment conditions (duct

height, atmosphere)

X

X

X

X

2.a Determination of optimal firing policies

14

Meshed engagement cells

threat

Ship 1 LR MEZ

Tree of permissible firing policies

Ship 2 LR MEZ

Ship 1 SR MEZ

Ship 2 SR MEZ

1. consider each cell for 1st shot

1.a. recursively, consider each cell and salvo size following assessment time

1.b. compute engagement timeline, PK, cost and add to possible firing policies

1.c. consider further shots if sufficient time

assessment time

Ship positioning, threat placement

S11 L S24 PK , Cost

earliest launch time for follow-on shot

cells ordered in increasing time

11

24

HVU

launch

intercept Given 1st shot

Evaluation of firing policy tree to determine the Pareto Efficient Boundary (PEB) for PK vs. Cost

15

232876 nodes (firing policies)

138027 nodes

2792 nodes

Complete tree Leaf pruning Local pruning

Pareto dominance condition applied during tree construction to prune leaf nodes

PEB determined after construction of complete tree

Pareto dominance condition applied locally during tree construction to prune branch nodes

Methodology • Construct firing policy tree

• Apply Pareto dominance condition at local branch level to confine tree growth

• Parse tree to determine PEB • Recursive algorithm to identify dominant firing policies

Graphs show Pareto Efficient Boundaries (PEB) for illustrative results, where solution points below and to the right are dominated by solutions on the boundary

Optimal firing policy to achieve a desired PK : Solution for TG configuration 1

16

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8

Kill

Pro

bab

ility

Mean Cost

bearing: 00

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8

Kill

Pro

bab

ility

Mean Cost

bearing: 0

bearing: 10

bearing: 20

bearing: 30

0

0

0

0

required PKTOT = 0.95

resulting cost C = 0.941 firing policy s2Ls2Ls1s1

0

0.5

1

1.5

2

2.5

3

3.5

0 10 20 30 40 50 60 70 80 90

Me

an C

ost

Target Bearing

(0,1),(0,8)

TG configuration 1 S1: (0,1), S2: (0,8)

300

C = 3.166 S1S1S1Ls1s1

100

C = 1.420 s2LS1Ls1s1

target bearing

1

2

1

(S2 1xSR) Look (S2 1xSR) Look (S1 2xSR)

Optimal firing policy to achieve a desired PK: Solution for TG configuration 2

17

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8

Kill

Pro

bab

ility

Mean Cost

bearing: 00

required PKTOT = 0.95

resulting cost C = 3.166 firing policy S2S2S2Ls2s2

0

0.5

1

1.5

2

2.5

3

3.5

0 10 20 30 40 50 60 70 80 90

Me

an C

ost

Target Bearing

(-1,1),(1,1)

(0,1),(0,8)200

C = 3.082 S2S2LS2S2S2S2

300

C = 1.719 S2Ls2s2LS2S2

TG configuration 2 S1: (-1,1), S2: (1,1)

target bearing

1 2

1

(S2 3xLR) Look (S2 2xSR)

SSM showing illustrative threat raid engagement

18

Metrics for main threat axis of -200

Launch and missile support schedule

PEB for T1 Cost as function of main threat axis angle

SSM can rapidly assess a single raid presentation (approx. 0.3 s for this scenario)

2.b Assessing TG configurations

19

weighted mean cost •narrow: Cnarrow = 3.10 •wide: Cwide = 2.44

0

0.2

0.4

0.6

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

“narrow”

0

0.05

0.1

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

“wide”

1 2

1

0

0.5

1

1.5

2

2.5

3

3.5

0 10 20 30 40 50 60 70 80 90

Me

an C

ost

Target Bearing

(0,1),(0,8)

weighted mean cost •narrow: Cnarrow = 1.34 •wide: Cwide = 2.53

0

0.2

0.4

0.6

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

“narrow”

0

0.05

0.1

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

“wide”

1

2

1

TG conf. 2

TG conf. 1

0

0.5

1

1.5

2

2.5

3

3.5

0 10 20 30 40 50 60 70 80 90

Me

an C

ost

Target Bearing

(-1,1),(1,1)

(0,1),(0,8)

preferred TG configuration given threat axis assumption

threat axis assumptions

cost – target bearing results

Illustrative SSM results for 3 TG configurations

20

1_188 17_409 22_273

Threat Axis Distributions (TAD)

TAD 1

“narrow”

TAD 2

“wide”

TAD 3

“flat”

2.b Determining the optimal TG configuration to minimise missile expenditure

21

• Heat maps used to identify optimal ship locations given assumptions on the main threat axis

• Permits what-if analysis, e.g. if one ship’s location is fixed • Production requires extensive number of TG

configurations to be considered

TAD 1

“narrow”

TAD 2

“wide”

TAD 3

“flat”

What-if analysis: Revised heat map given S1‘s location fixed, TAD 1

XX

X

Summary: TG AD study 1. MEZ construction using physics-based simulations

22

storage processing

servers

1. model & data deployment

3. data storage

2. model execution

1000+ processors

4. result queries for MEZ construction

MEZs

• 1 to 2 days for HTC model execution • 1 week turn-around, including data

parsing and post-analysis

Air Defence Model 6 DOF Missile Model

Each MEZ captures radar detection, tracking, SAM fly-out, intercept and endgame aspects of an engagement as represented in the physics-based simulations

storage processing

servers

1. model & data deployment

3. data storage

2. model execution

1000+ processors

Ship Stationing Model Command Line Interface

Summary: TG AD study 2. TG ship stationing analysis using MEZs

23

FP: T1: S1S1 T2: S1S1 T3: S2LS1 T4: S2LS1 C = 6.6 PK = 0.9

1

2 3 1

2 3 4

XX

X

MEZs

TG configuration results

• 1 hour for HTC model execution of 105 configurations • 1 day turn-around for data storage and post-analysis

HTC permits extensive search of parameter spaces to support optimisation problems

Number of TG configurations: NTG = (11 x 21) x (21 x 21) x (21 x 21) ≈ 45x10

6

Summary: TG AD study 3. TG aggregated results

24

storage processing

servers

model execution

1000+ processors

TG configuration results

heat maps

Ship Stationing Model Graphical User Interface

FP: T1: S1S1 T2: S1S1 T3: S2LS1 T4: S2LS1 C = 6.6 PK = 0.9

1

2 3 1

2 3 4

0

0.05

0.1

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

0

0.2

0.4

0.6

-90 -60 -30 0 30 60 90

Pro

bab

ility

Target Bearing

Cmn = 5.8 Cmn = 6.2

optimal TG configurations

threat axis assumptions

A simplified model, layered over and underpinned by physics-based simulations, can extend analysis to larger parameter spaces

Concluding remarks (1 of 2)

Two layered approach, OA model underpinned by physics-based simulations, permits exploration of a large parameter space in a structured manner

Analysis benefits being realised through HTC

Permits high fidelity Monte Carlo simulations to be used in a wider context

Permits brute-force extensive search of large parameter spaces, but parameter space increases can still outgrow HTC capacity

Data management challenges arise from the vast amounts of data that can be rapidly generated

M&S fully embraced the approach to construct MEZs

Gives firm basis and credibility to the ship stationing analysis

Approach adopted to support planning for an upcoming trial

Exploring feasibility of peeling back the layers in a MEZ to separate out detection, tracking and SAM fly-out

25

Concluding remarks (2 of 2)

OA model observations and benefits

In-house development responsive to evolving requirements

Fast running thereby permitting large number of cases to be quickly considered

Push to increase functionality:

Visualisation of MEZs from the physics-based simulations

Command Line Interface to utilise HTC, Graphical User Interface to visualise results

Data output control to minimise amount of data produced – better to quickly run large number of cases with minimal data recording and then rerun the cases of interest

Heat maps for visualising TG ship stationing results

Further work:

Validation and focused runs using physics-based simulations

Increase complexity of engagements represented

Explore exploitation of the engagement planner

Explore exploitation of the heat maps to support tactical planning

26

Questions ?

27

© Her Majesty the Queen in Right of Canada (Department of National Defence), 2019