By Michael Webb, PhD, SPA Software Development Fellow
SPA has supported multiple Motor Aging and Surveillance Technology projects for the Air Force Research Laboratory at Edwards Air Force Base. Our most recent effort focused on developing technologies to meet the Rocket Propulsion of the 21st Century (RP21) Aging and Surveillance goals. These goals include physics-based models, nondestructive evaluation (NDE), enhanced NDE systems, and solid rocket motor sensor technologies. RP21’s main goal is to reduce predictive uncertainty of the service-life estimate of rocket motors.
To help achieve this goal, SPA established a team of experts to design and demonstrate the Missile Inventory Management System (MIMS). Historically, rocket motor programs have managed the fleet as a whole, without analyzing motors individually based on their unique history. Using MIMS, fleet managers can model the service life of individual rocket motors, examining their manufacturing history and the effects of exposure to unique working environments. For example, environmental factors such as temperature and humidity can affect a rocket motor’s service life dramatically. Understanding fleet health at the individual rocket motor level is key to reducing predictive uncertainty.
MIMs combines advanced, physics-based models with environmental and manufacturing data for individual motors in a cloud-based architecture. The system allows motor fleet managers to evaluate each motor on all timescales of interest, providing short-, medium-, and long-term estimates of each motor’s service life.
MIMS incorporates both cutting-edge and established technologies as well as advanced physics-based models for predicting rocket motor service life. The system is built in layers: a database layer based on PostgreSQL; a processing layer that parcels calculations to workers as instances on Amazon Web Services (AWS) Elastic Compute Cloud; a set of networked sensors that continually gather environmental data and send it to the MIMS database; and a customer-facing layer for users to see individual rocket motor data, service-life predictions, and rolled-up information about the fleet as a whole.
MIMS needed to ingest large amounts of legacy data as well as process new data hourly. Using AWS, the team created a unique blend of server-less applications and custom virtual machines. The configuration includes a control server and a flexible cloud of worker nodes, each taking a bit of the load. Although setting up the communications and control flow was highly challenging, the SPA team enjoyed implementing the new technologies.
The processing layer executes a wide variety of workloads, ranging from sitting idle to running large computations that would take months to complete on a single processor core. To handle such a large range of workloads quickly and efficiently, the team designed the processing layer to be able to scale up more compute resources. These resources are not necessarily on the same machine or even on the same network. The processing layer uses Dask (a Python distributed computing package) to orchestrate the computations and data exchange among this flexible number of compute nodes. Dask proved to be an excellent tool for this purpose. The requirement for a distributed system posed unique problems that required close collaboration to ensure the processing layer’s compatibility with each layer in the system.
The sensor layer collects and automatically processes weather and environmental data from custom-built, multisensor platforms as well as online weather services. It also obtains bulk data from client test sites. Each source is uniquely formatted and requires pre-processing before it becomes a part of the MIMS database.
The SPA team developed multiple statistical and mathematical models to provide insights and predictions into the physical structure and chemical state of a rocket motor at any point in the future. For example, the team implemented a physics-based diffusion-reaction model that captures changes in chemical composition of the motor materials as the environment changes. This computationally intensive system required several optimization techniques across the disciplines of mathematics, physics, and software engineering.
The customer-facing layer links the users to the powerful predictive machinery of the MIMS model and databases. Custom visualizations were implemented in vue.js, connected to the PostgreSQL database through REST endpoints. The large amount of data available to users required the development team to design unique, dynamic webpages that responded rapidly to the user’s requests for data and information.
SPA built a highly effective team to work on this challenging project. Whether it was assembling and configuring sensors with RaspberryPi, setting up a cloud-based database, building a web-based User Interface, or solving differential equations on machines with 48 CPUs, each person on the team contributed the expertise and knowledge needed to make a significant contribution.
A process of predicting future costs based on past experience using today’s knowledge and expertise.
This article, by Trip Barber, SPA Chief Analyst, was originally published in the MOR Journal, June 2021.
Decisions are complex. What if you could model and explore the effects of any decision before you implement it in the real world?