Photo By Jhon Parsons | Minh Vu (left), SPY-1 engineer at Naval Surface Warfare Center, Port Hueneme Division...... read more read more
Subscribe
A naval engineer and a university professor have harnessed machine learning and data automation to help predict when AN/SPY-1 radars need maintenance and minimize downtime.
The collaboration between Minh Vu, a SPY-1 engineer at Naval Surface Warfare Center, Port Hueneme Division (NSWC PHD) in California, and Jason Isaacs, a computer science professor at California State University Channel Islands (CSUCI) in Camarillo, aims to boost radar readiness in the Navy fleet.
The project is funded by the Naval Innovative Science and Engineering program.
According to Vu and Isaacs, preliminary results have shown that the predictive model they developed can forecast when the SPY-1’s cross-field amplifier (CFA) channels need maintenance at least one test cycle in advance.
The project also promises to save sailors time by avoiding unnecessary routine maintenance when the SPY-1 equipment is in good operating condition, and by using automation to eliminate the manual data entry involved in monitoring the health of the radar system.
By automating data collection, the project ties in with NSWC PHD’s digital transformation goals. The effort also leverages partnering with academia, according to Chief Technology Officer Greg DeVogel.
“It’s a big help to our command to bring in someone with expertise in an area like artificial intelligence,” DeVogel said.
Vu credited DeVogel and Alan Jaeger, who manages research and technology applications for the command, with sparking the partnership for the SPY-1 project.
“Greg and Alan had the great vision to bring Professor Isaacs onboard,” Vu said.
The project supports NSWC PHD’s role as In-Service Engineering Agent (ISEA) for SPY-1 — a critical component of many Navy warships’ combat systems.
Supporting SPY-1
The SPY-1 radar integrates with the Aegis Weapon System to track, detect and engage threats and provide ballistic missile defense capability. The radar employs four antenna arrays, which are large octagonal panels positioned around a ship’s superstructure for 360-degree coverage.
One of the top indicators of a SPY-1 system’s readiness is its effective transmit power, or ETP, which measures how much power the system puts out for each array. Another test metric is the isolation values, which measure power to a designated array when the others are isolated from power.
When ships are underway, sailors routinely run tests on their SPY-1 systems and manually record the ETP and isolation values to submit to NSWC PHD.
“They capture that data and manually enter those two parameters into a spreadsheet,” Vu explained. “The test itself has many more important parameters, but manual data entry of those has not been feasible.”
To streamline the process while capturing those additional test parameters, Vu sought a way to automate the SPY-1 data collection. He turned to two software programs that other warfare centers have developed.
“These programs allow the ship to automate the data capture of the test results we want and then send it ashore via communication systems,” Vu said. “This data is used to indicate the health of SPY-1 radar transmitters and provide situational awareness to leadership decision-makers.”
NSWC PHD receives the data and sends out reports to program office leadership, type commanders and other Navy officials.
The additional test parameters that the data automation collects will help ISEA personnel and Regional Maintenance Centers better assess the condition of SPY-1 radars and bolster fleet support, according to Vu.
Along with avoiding manual data entry, Vu and Isaacs’ project aims to boost efficiency by enabling sailors to employ predictive maintenance. As opposed to preventive maintenance, which is typically scheduled based on set time intervals, predictive maintenance is performed only when needed based on the condition of equipment.
The predictive maintenance aspect of the project relies on machine learning, which is where Isaacs’ expertise comes in. The CSUCI professor is an authority on artificial intelligence, and he had previously worked with NSWC PHD on efforts involving robots.
Supervised learning
Isaacs teaches courses in computer science, information technology and mechatronics engineering at CSUCI.
Mechatronics is a multidisciplinary branch of engineering that includes robotics. In that realm, Isaacs has coached CSUCI students competing in NSWC PHD’s Robot Rodeo at Fathomwerx Lab in the Port of Hueneme.
In 2020, Isaacs joined NSWC PHD’s first summer faculty program, a 10-week fellowship between naval labs and research experts that the Navy’s Office of Naval Research facilitates. Working with NSWC PHD that summer, Isaacs developed algorithms to control the movement of robots and autonomous vehicles.
About two years later, Isaacs reconnected with DeVogel at the first Robot Rodeo in 2022. They discussed several ideas for projects that Isaacs could work on in another summer faculty program at the command. One of DeVogel’s suggestions was the SPY-1 readiness project with Vu.
That summer, Isaacs and Vu evaluated different types of machine learning that could potentially help in predicting the need for radar maintenance based on data samples from CFA channel amplitude measurements. They narrowed in on a category of machine learning called supervised learning, which uses labeled sets of data to teach an algorithm to identify patterns and predict outcomes.
“Basically, what you’re doing is taking examples of what the data would look like at different points in time,” Isaacs explained.
To illustrate the concept, the professor described a scenario in which there is a sample of data at times 1, 2, 3 and 4.
“We could give the algorithm the first three time samples and the value that should have been predicted at time 4,” Isaacs said. “With enough examples, we can take away the fourth and have (the algorithm) predict the fourth, or the fifth or sixth.”
As developers feed more data into an algorithm and prompt it to make predictions, they can tell the algorithm when it’s right or wrong — in other words, supervising its learning or training it.
“Training the algorithm is very time consuming, but then the act of prediction can be done very quickly,” Isaacs said.
Vu and Isaacs said they made good progress on developing their predictive model during Isaacs’ summer faculty program stint in 2022. Since then, Isaacs has continued working with Vu at NSWC PHD part time as a contractor.
The partners initially trained their algorithm with a simulated data set of CFA channel amplitude measurements, but they have since shifted to real data from the fleet.
Preparing to launch
The project is now in a phase of validating the algorithm based on actual data from ships’ SPY-1 radars. That will help Vu and Isaacs refine their predictive model before putting it into service.
“Once we’re confident with the results, we would like to either deploy it to a cloud-based server or a readiness dashboard,” Vu said.
After deploying the model, NSWC PHD’s subject matter experts (SMEs) could remotely access a variety of test data from a ship to help them monitor its SPY-1 radar.
“Automating the data capture will help us, the SMEs, to better assess the system than before, when only having the ETP number and the isolation values,” Vu said.
He added that the automation will also save sailors time and eliminate errors that can result from collecting data manually — again supporting digital transformation and fleet readiness. The automation process has been successfully tested aboard numerous ships and land-based Aegis Ashore test sites.
| Date Taken: | 08.08.2024 |
| Date Posted: | 08.09.2024 15:00 |
| Story ID: | 478133 |
| Location: | PORT HUENEME, CALIFORNIA, US |
| Web Views: | 50 |
| Downloads: | 0 |
This work, Partnership With Professor Produces Model to Predict SPY-1 Radar Maintenance Needs for Navy Fleet, by Thomas McMahon, identified by DVIDS, must comply with the restrictions shown on https://www.dvidshub.net/about/copyright.
No keywords found.
