AI Heads to Space with a Constellation of Use Cases

Artificial Intelligence (AI) is one of the latest buzz terms for the satellite industry, and it’s clear that it could have a dramatic role when it comes to many aspects of the ecosystem, including system manufacturing, in-orbit management and delivering value to applications. Although it’s still early days for AI, and there are challenges and gating factors to be hurdled before the technology sees mainstream deployment, its potential in space is rapidly developing.

The pop culture view on AI tends to revolve around the twin tropes of the evil digital overlord (HAL, SkyNet in “The Terminator,” “The Matrix,” et al), and friendly digital assistant-style interactive helpers (Siri, Alexa, “Her”). The reality is that in industrial use cases, AI initially has applicability not in replacing higher-value human activity (let alone bringing sentience to bear), but rather in supplementing it. In other words, it can swoop in and take over what can best be described as the busywork of the industry. This is nowhere more evident than in the satellite manufacturing space.

“We are trying to use human intelligence effectively and that’s difficult enough — but AI digitalization could be used to support our engineers and other people in the organization to do that in better ways,” says Andreas Lindenthal, Chief Operating Officer (COO) at satellite manufacturer OHB Systems.

For example, there are hundreds of thousands of electronic piece-parts that are used for building various larger systems within satellites. In the manufacturing supply chain, thousands of parts are required every day, and they all need to be de-stored from their controlled holding environment, checked to see if they’re fit for purpose, delivered to the floor and then tested to see if they’re performing as expected within the manufacturing process. That’s a labor-intensive and complex process for humans to manage — but an AI could automate the visual checks and inventory management to enable an optimized approach, at scale, in a fraction of the time it would take a human team.

Similarly, AI could be particularly useful for maintaining data consistency in the engineering process. Manufacturers manage hundreds of documents that describe a satellite’s design, and how that design is verified and validated.

“We are tasked with maintaining consistency in terminology and approach across the various documentation pieces detailing the process and procedures to build a satellite,” explains Lindenthal. “We must make sure that everything is correct and consistent throughout the documents, as well as ensure that the data used for verifying the performance of the hardware is standardized. AI could easily replace the human effort in managing and monitoring for that kind of consistency and data.”

There are yet other use cases for the factory floor; for example, AI could examine image captures of live processes — say, the creation of circuit boards — to determine whether a box has a defect or anomaly.

“In our business, visual inspection is very important,” notes Jean-Philippe Jahier, deployment director for innovation and new technologies at Thales Alenia Space. “When you want to check your work, you look at it and compare what you see with what you expect, to determine any physical defects on material. If you can use images of manufactured parts to automate that, it’s potentially a large source of optimization.”

AI could also observe the manufacturing equipment itself, via video feeds or telemetry data, to determine whether a machine is functioning as it should and to perform predictive maintenance by anticipating when performance thresholds are being reached. It then becomes possible to replace any faulty equipment before failures occur.

It’s important to note that AI may represent enormous possibilities for process improvement, but its deployment is not without challenges. One of the main issues on the manufacturing front is the intense amount of customization that’s required. No two satellites are identical, despite attempts to standardize products and systems.

In practice, this means that developing the manufacturing and verification process for a specific satellite tends to be a tailored affair requiring plenty of interaction with engineers. AI thrives on repeatable processes, not only to automate them, but also to learn to identify anomalies, which it typically does by allowing (and learning from) a certain amount of failures. However, this in turn introduces risk into supply-chain proceedings.

“It’s quite easy to inspect clones of the same object, but you need more maturity in the algorithms to catch flaws in a variant object,” says Jahier. “Satellites all have different wiring or terminals with variables in size, shape, color or the number of wires. Better and smarter AI is required to detect defects in this scenario directly from images.”

Naturally, automation and the speed at which machines can perform their tasks are the main benefits that AI can provide in many scenarios, along with the fact that it never gets tired, bored or angry about work-life balance. That means that AI can also help improve time to market by working tirelessly round-the-clock.

A prominent example of this is the development of fully automated test environments. At OHB for instance, customers expect 100 percent verification of all the possible operational modes that could occur in orbit. That’s a process that requires tens of thousands of payload and environmental tests. AI could perform the tests more quickly, and could also learn to sequence them for more efficient, full optimization of the entire test gamut.

“The systems that we have in place run for several days and nights, just switching from one mode to the next to check if the deliverables perform as expected,” says Lindenthal. “Once one mode is verified, we switch to the next one. AI could be crucial for optimizing and reducing the time of each of the test phases, perhaps by as much as 50 percent.”

AI could also improve time to market in the design phase by acting as a kind of peer review that monitors performance indicators and best practices.

“Imagine a system with the capability to analyze the output of the designer and check for inconsistencies,” explains Jahier. “Large teams of designers are always exchanging emails. AI could monitor this and raise concerns if an issue stays unsolved for too long a time; it could then raise the attention of management. It’s a hypothetical, but a good example of a design oversight use case.”

Manufacturing isn’t the only sector of the space industry where AI could overhaul legacy processes. AI also has a big role to play in upstream in-orbit apps, including service maintenance, refueling and repairs.

“The general approach now is to send a satellite up there to do the manipulations required for these kinds of tasks, but they’re controlled from Earth by humans,” says Shagun Sachdeva, analyst at Northern Sky Research (NSR). “AI could automate that process and manipulate the robotics within the context of anything else that’s happening, such as dealing with space junk, without a human overseeing making that happen.”

One example of this working in practice comes from SES, whose fleet of 70 satellites each generates continuous health and status data. The operator is working with IBM Watson to use a real-time streaming AI to look at that data lake and increase reaction time to correct and normalize satellite operations when fluctuations in the telemetry occur.

“The traditional approach is very manpower-intensive and makes it difficult to correlate telemetry with other trends or events, such as a solar storm or space debris,” says Ruy Pinto, Chief Information Officer (CIO) at SES. “We are an industry that’s growing and being disrupted by new technologies and new entrants, and it’s vital for the larger satellite operators to look for ways of reacting quickly to customer demands, and to look for safer and more efficient ways to operate our fleets. Everyone is looking at ways to improve autonomy and doing a better job of serving customers.”

Earth Observation (EO) is a classic — if not the most classic — use of satellite deployments; the images generated from above can be used for any range of use cases, from understanding the effects of natural disasters to carrying out espionage efforts. AI however opens up a new chapter in EO by lending scale, automated image tagging and sorting, and map correlations — which, taken together, can be used for entirely new applications, such as understanding the economic state of a third-world nation.

Radiant Solutions, the analytics arm of satellite manufacturer Maxar Technologies, recently signed a contract with the U.S. National Geospatial-Intelligence Agency (NGA) to provide more than 1 million labeled objects within high-resolution satellite images that will be used to accelerate the development of machine learning algorithms that can extract valuable information from imagery at scale. Ultimately, that information will be used to support national security and humanitarian missions. By training algorithms to automatically identify dozens of objects across 60 object classifications — such as damaged buildings, construction equipment and tents — it becomes possible to create maps and correlations that can be used to gain insights into pressing global challenges, such as stopping the spread of disease and improving infrastructure in the developing world.

“We look at a current image, count everything, overlay with a base map, and add the new features to the map,” says Adam Estrada, director of software development at Radiant. “Our constellation revisits the same areas every day or so, and we’re able to see spikes and dips in activity.”

AI makes it possible to do this at scale, quickly: The algorithms take 0.3 seconds to scan and tag objects in a one square kilometer area versus 25 minutes for a human.

“The performance is 128 times faster than a person doing the same task, which means we have the ability to do things like rapid damage assessments,” says Todd Bakestow, head of the SpaceNet program at Radiant and director of strategic alliances. “Think about Western Africa with Ebola outbreak. This information could have been used to direct resources, estimate the population by counting buildings, and help workers know where passable roads are. When we think about imagery analytics, humans are pretty good at the accuracy piece already. Speed and scale is where the automation really helps. That is the true power of this.”

SpaceKnow is a startup with a similar mission. Its AI scans the world, tirelessly counting cars, trucks, roads, buildings, aircraft and other kind of objects, on a massive scale, for entire countries. Asset management companies then use that data to manage portfolios.

“We can tell our hedge fund and large financial clients about activity at a specific company’s mining site, factory or port — or give a comprehensive evaluation on the state of an entire economy,” says Pavel Machalek, co-founder at SpaceKnow. “For instance, we measure the GDP of all the provinces in China on a weekly basis, by counting objects and classifying materials, and observing industrial production and other activity. We determined the Nigerian GDP uptick way before everyone else by looking at the manufacturing conditions across 6,000 factories. It’s econometrics — a new approach.”

Eventually, AI will be able to count and identify every object, everywhere, every day. Machalek said that he can envision a live trigger warning system for the entire world, where AI will monitor everything in full autonomous mode and notify the humans when something interesting happens, such as a new building being constructed in the past week. This has applicability beyond high finance, of course.

“It’s an opportunity to reveal the full complexity and beauty of the Earth, all 7 billion people’s environments at once,” he said. “We will be able to offer better city planning, and better manage water, and natural and industrial resources, and save human lives.”

Vertical industries are eyeing AI with interest, particularly when it comes to emerging technologies. For instance, in the transportation sector, satellite information and capabilities can be combined with Internet of Things (IOT) sensor data to enable streamlined operations for shipping and airline activity.

“A lot of data is coming online and being focused on Machine-to-Machine (M2M) communications and IOT,” says Dallas Kasaboski, analyst with NSR. “People are digitizing anything they can. AI in transport applications is enabling trends to be extracted from a large and growing variety of data, to enable better business decisions.”

For instance, a merchant ship with hundreds of containers may have sensors installed on the engine and navigation systems, monitoring how much fuel is being consumed and helping to optimize navigation. There may also be sensors on the cargo (there are, for instance, new regulations calling for produce being shipped at a specific temperature).

“You take that info, automate it and use AI to do predictive modeling,” Kasboski notes. “Using satellite data, you can combine the telemetry with weather charts and forecasts, and port activity and more, to automatically determine the best path to conserve fuel or shorten the journey. The same thing is also happening in land transport, railways and so on.”

Datacomm operator SES meanwhile is designing its next-generation capacity systems to be autonomous, with a dynamic resource manager that will be capable of making decisions and adjusting coverage, capacity and spectrum as needed. This has great applicability to the transport sector.

For instance, airline requirements for satellite coverage vary by time of day depending on how busy an airport or airspace is. The AI can look at the trends and traffic patterns of the data being exchanged with a plane, correlate that with Key Performance Indicators (KPIs) to determine if performance thresholds are close losing compliance with Service-Level Agreements (SLAs), and make a decision on how to alleviate the issue via smart caching, providing more capacity, traffic shaping and so on. In an IOT scenario, this dynamic adjustment — based on context, not just a set of rules and policy — can become invaluable.

“This type of use case lends itself to the autonomous side. Not only can we provide dynamic resources as needed, but an AI can also learn from the traffic pattern how best to serve that location, based on priorities provided by a human team,” explains Pinto. “You need an AI component in the central part of the network that knows how it can optimize the user experience.” VS