tatiana kichkaylo with dave barnhart and lucy hoag usc information sciences institute
TRANSCRIPT
Tatiana Kichkaylowith Dave Barnhart and Lucy Hoag
USC Information Sciences Institute
BackgroundFunded as ISI Inter-divisional seedling
Space Engineering Research Center+Artificial Intelligence
= Spacecraft Portal for Integrated Design in Real time
Driven by user needs SERC is building satellites, fast, cheap, with student power
("teaching hospital" model) SPIDR is there to help
Systems Platform for Integrated Design in Real time
Design problem Spacecraft missions are typically “one-off” designs, requiring high degree
of non-recurring engineering between even similar design states Spacecraft design is highly interdisciplinary, and follows many sets of
constraints and rules dictated by environment, payload and mission sets
Components and design approaches can be reused Engineering knowledge can be encoded to automate many design choices
and analysis steps "Paperless" design helps directly and also provides basis for further
automation
Ultimately want to let individual engineers bounce ideas of a computer system, leading to faster, cheaper and better designs
BlueprintsPurchase ordersLaunch dates…
BlueprintsPurchase ordersLaunch dates…
Magic
BlueprintsPurchase ordersLaunch dates…
Design
Geometry/mass
Power
Thermal
Launch
Orbit
Flight software
Mission goals
Vibration
Component providersExport regulations
Delivery schedules
Manufacturing capabilities
Validation and testing(software and hardware)
Payload
What does it take to build a spacecraft?
Bigger pictureCollaborative design
Information sharing and clash detectionOptimization of parameters
Limited number of parameters but with high fidelityProject management
Schedules and resources
SPIDR Automate roll out and propagation of dependencies Optimize over structures given high-level specification Very lightweight and driven by project needs
Idea: Use planningAI planning is about finding a sequence (DAG) of
actions that achieve a goal Methods have preconditions Methods expand parts of the plan
Design is about finding a set of recipes that convert a specification into a complete design Recipes have applicability conditions Recipes modify (expand) design
Both methods and recipes have parametersBoth plans and designs have constraints
Constraints in design are often represented as models
User revises requirements and metrics as needed
Reusable knowledge base of engineering
rules
Mission requirement
s, orbit
N best designs(component lists and
system budgets)
Reusable component database
Optimization metrics
SPIDR
Reusable data
Mission-specific data
Automated design synthesis
Design loop
Rule example
Semantic entities
Query engine
Variables and
constraints
Constraint propagation
Tokens and rules
Planner
Semantic layer
Constraint layer
Planning layer
Data representation Algorithms
Optimization engine
User interface
iLan
ds
SP
IDR
Mission req's
Constants
P. battery
Implementation
SPIDR meets real world
Nanosatellite project at USC ISI Low costs Unusual constraints
Reality alters trajectory Initial "vision" for SPIDR
Single optimization tool with models in the loop Actual use
Run every several days with design files (XML etc.) posted online
Got several extensions, model tweaked constantly Useful for dealing with ripple effects and ensuring consistency Users want more
ObservationsNo way to close all loops
Extended to dynamic models (e.g., power profile) Stuck on thermal and geometry due to
Lack of APIs Licensing issues Need for user interaction
One shot is not enough – need Living Design Feed back from out-of-the-loop models, version control Side-by-side designs Project structure and scheduling roll out Still want the optimization loop
Observations (2)Rule design is not easy
Engineers understand the concept Tools help: Eclipse plug-in
Crossing domain boundary is hard Tweaks and overrides encourage local thinking SPIDR Lite to study the knowledge engineering problem
Other applications:Trade space studies and tools for acquisition SPIDR-Energy for undersea surveillance (DARPA)
Last slideSPIDR is about rolling out engineering recipes and
optimizing over high-level structures Glue for collaborative design, optimization, and project
managementSPIDR is driven by real project(s)
Some ambitions got scaled down New extensions and ideas emerged
Next steps New [engineering] domains New applications More users (students?)
Tatiana [email protected]