This article is published in Aviation Week & Space Technology and is free to read until Oct 23, 2024. If you want to read more articles from this publication, please click the link to subscribe.
Governing data collection processes will be crucial to ensuring AI systems can yield good results.
The advent of artificial intelligence is forcing operators to revisit data platforms and data quality—including input sources, information processing and validation—to ensure those systems add value to their maintenance planning.
Artificial intelligence (AI)-driven solutions usually start with the integration of those new tools into existing infrastructures. MROs and technology developers are seeing increasing opportunities for applications in aircraft maintenance planning.
Rob Mather, vice president of aerospace and defense industries at enterprise software provider IFS, identifies two main routes for introducing AI into the IT ecosystem: standalone systems or those already integrated into enterprise resource planning (ERP) or MRO software.
Stand-alone solutions could be bespoke or off-the-shelf. “In this mode, the AI typically sits on top of the data infrastructure, such as a data lake, for instance, but not integrated with operational systems,” Mather says. “This stage can provide a decent level of insight, but implementing them would require additional steps.” The second approach deploys systems that are already integrated into ERP or MRO software—for example, implementing AI-driven insights directly into the operational systems, he explains.
Saravanan Rajarajan, director of aviation solution consulting at Ramco Systems, suggests that companies first evaluate the data capabilities and quality of the source systems. “Existing infrastructure should have a clear process and workflow to collect the right data, and collected data should be governed by the right workflow controls,” he says.
Rajarajan also highlights the importance of training AI systems to distinguish different types of data and their context, as well as relationships among structured and unstructured datasets. “Accuracy issues require constant intervention from the data engineers and business until the AI systems bring in real value,” he says.
MODELING AND SIMULATION
MROs can use AI modeling and simulation techniques to make informed decisions and to predict outcomes of maintenance strategies.
Simon Miles, head of AI at startup Aerogility, says model-based AI can help aircraft operators create holistic, accurate digital portrayals of their entire operation. Each key element—a component, a facility or a person—is represented as an agent and configured with specific characteristics. Each individual agent operates and interacts with others over a period of time to simulate the operation.
“This provides an enterprise digital twin, allowing diverse stakeholders to see different views on the same virtual model and make strategic decisions that are complementary and coordinated,” Miles explains.
Simulation tool experts at engineering software developer Ansys say AI’s ability to build reduced-order models quickly offers the biggest value to MROs. These models can be probed in real time to investigate the impact of a particular flight cycle. “It is possible to evaluate the impact on materials and better plan maintenance schedules or reengineer components to extend the amount of time between required overhauls,” says Mike Slack, Ansys senior technical account manager.
Meanwhile, Aerogility says its AI capability helps aircraft operators manage complex data and knowledge, augmenting both long-term strategic and immediate tactical decisions. According to the company, well-engineered AI software should be able to accommodate larger quantities and varieties of perspectives because it can model and gauge how organizational demands interact rather than just considering each in isolation. For example, planning to maximize aircraft availability or maintenance yield depends on supply chains, budgeting constraints, staff availability and management, among other factors.
“All aircraft maintenance checks can be modeled within Aerogility while accounting for multiple utilization drivers,” Miles says. If a C check is due on Day 730 at 7,500 flying hours or after 5,000 cycles, for instance, the program will schedule the check when the aircraft reaches one of these limits, he says. Additionally, the program can merge checks that are within close proximity to one another.
ADVANCING PREDICTIVE CAPABILITIES
Identifying failures before they occur has benefited aircraft operators in cost savings and operational efficiency. Predictive maintenance is not new, but it has evolved over the years from the big data concepts of the early 2000s to the emergence of machine learning in the mid-2010s.
In parallel, the development of connected aircraft, such as the Boeing 787 and Airbus A380, brought terabytes of new data with every flight. Mather from IFS says the ability to generate and interpret real-time data became more widely available around the start of the 2020s. “But you still needed those pricey and rare data scientists to label all the data and guide the learning,” he notes.
Mather suggests the industry is at an “unsupervised learning” stage. “It sounds scary, but it just means we don’t need to label the data, and that’s a much faster and less expensive proposition,” he says.
Mather asserts that current predictive maintenance processes will still produce effective results, but using AI lowers costs and increases access. He says AI also enables anomaly detection models to discover previously undetectable correlations in data. “It doesn’t need to know why something happens; it’s just recognizing patterns,” Mather says. “All this drives a higher degree of accuracy in the predictions, but [AI is] really more about being available more widely.”
From a modeling perspective, AI lets users plan for large and small maintenance events—expected and unexpected—as well as the associated costs, and improve their understanding of the likelihood and severity of financial impact variables.
Miles at Aerogility says he thinks companies can use the technology to examine these issues, identify the most likely scenarios and make better business-critical decisions. For instance, if a forecast anticipates aircraft maintenance delays when specific spares are unavailable, users can plan and simulate the best response by modeling the impact of an expanding or shrinking spares pool to ensure operations can keep working smoothly.
Miles adds that many often overlooked financial aspects of an operation—such as negotiating new MRO contracts and visualizing multiple offerings—can be modeled to assess their impact on cost and ground time.
PREVENTING BAD DATA ENTRY
Mather stresses that data is the food on which AI grows; good data will yield good results and vice versa. He believes it is easier to correct the data at the point of entry using systems that record at a very granular level and help improve data quality by enforcing rules on and guiding the technician. “It makes it easier to get it right and hard to get it wrong,” he says.
Ramco’s Rajarajan advises that operators should audit data sources regularly to identify inconsistencies. “AI systems should have data filters to prevent malicious or nonstandard data from entering through any of the input sources,” he says. AI systems also should detect and flag outliers for elimination or manual assessment.
Unstructured data such as PDFs hold a wealth of information, but Rajarajan contends that their nonstandard nature poses risks to data quality and security. “The right tools and processes are essential to extract and validate the data before it is pushed to ERP and AI systems,” he says.
Ansys’ Slack recommends that users ensure they understand the source and context of the original data, which will be directly relevant to the tools they derive from it. For simulation-based AI, he stresses the importance of knowing the extent of the operating and design variations from which learned outcomes are drawn. “To this end, simulation and data management tools will be key to providing this context and enriching the value of a company’s stored data,” he says.
Moreover, tools derived from AI or machine learning can help refine modeling and in turn optimize the design of the engineering system or how it is managed in service, Slack adds.
