AI on the Ascent: Practical uses for AI in Aviation Maintenance

Home » AI for Aviation Maintenance

Written by Rob Mather, Vice President, Aerospace and Defense Industries at IFS

Since the emergence of ChatGPT, it seems that Artificial Intelligence (AI) is the predominant topic. However, beyond public attention, AI has the potential to open up broad prospects for improving efficiency and productivity in a variety of industries.

But how can we implement AI in sectors such as aviation, known for its focus on safety and traceability, when these requirements may not align with our traditional understanding of AI? AI goes far beyond generative AI such as ChatGPT and large language models (LLMs). The integration of AI in aviation presents a unique opportunity to improve safety, efficiency, and accuracy through a wide range of tools and technologies.

There are two possible paths for the future use of AI in aviation: use AI only to present options and support human decisions or adopt explainable AI.

Visualization of information for decision support. A fundamental principle in aviation is the idea that a trained and certified person takes responsibility for decisions, based on appropriate authority and licenses. AI can be used to enhance and assist rapid decision making by quickly providing relevant information and options and leaving the final decision to the authorized person. By using AI as a support tool, people can make decisions more quickly and accurately without compromising control.

Accountable AI must be transparent. An alternative in the use of AI in aviation is the concept of explainable AI. Explainable AI includes tools and frameworks that facilitate the understanding of predictions generated by an AI model. While using a model, such tools help to improve its accuracy. When analyzing the results of a model, they are useful in understanding the process that led to a specific conclusion or result.

Many algorithms derived from machine learning are considered “black boxes” that produce results in a seemingly magical way. This can set up as a problem in the aviation context, where actions must comply with regulatory requirements and the motivations behind those actions must be understood by regulators. Claiming that “AI magically decided” will not constitute an acceptable explanation. Ensuring that AI is explainable is critical to its widespread adoption in aviation.

Ensuring acceptance of AI use in the industry is one thing; achieving tangible benefits is another. Four practical ways in which AI can bring real value to aviation maintenance come immediately to mind:

1.Maintenance Planning and Supply Chain Optimization:

There are several types of Artificial Intelligence (AI) models capable of performing optimizations, but some of the most common follow iterative approaches, running hundreds of thousands of scenarios instantaneously and regressing the results toward an optimum. Although AI-based optimization engines do not enjoy the same emphasis in public debate as generative AI, there are numerous applications for optimization in aviation. Focusing on aviation maintenance, optimization of maintenance scheduling emerges as the most obvious. Giving up maintenance efficiency means performing more maintenance operations over time, generating additional costs. However, extending the period when an aircraft is out of service means lost revenue. An optimization engine that can schedule maintenance at the most appropriate time and place has the potential to significantly lower maintenance costs and improve maintenance performance across the entire fleet.

At the same time, optimizing the order in which tasks are performed and assigned is essential to ensure the efficient and timely completion of a specific maintenance visit. This offers significant benefits in increasing process efficiency, reducing costs, and improving Technical Activity Time (TAT). This will help get the aircraft back in the air faster, generating increased revenue.

In addition, by applying optimization to the maintenance supply chain, material delays can be minimized, ensuring that the right components are always available where and when needed to enable the fastest return of aircraft to flight.

  1. Error detection and reclassification

Artificial intelligence can play a crucial role in detecting data entry errors and reclassifying data to improve the accuracy and overall quality of datasets.

Misclassification of faults in the ATA system is a widespread issue in the aviation industry. Sometimes, this is due to human error, but frequently the fault is categorized at the time of reporting, based solely on symptoms. Later, when the problem is resolved, it is discovered that the system initially identified was not the real culprit, but that the root cause lay elsewhere. These misclassifications can significantly impact data quality. Airlines may rely on the technical records team or reliability engineers to review records, identify errors, and reclassify them to resolve the issue. However, this process can be lengthy, time-consuming and require painstaking attention to detail.

Southwest Airlines, a client of IFS, recently implemented an artificial intelligence-based solution capable of detecting misclassified faults by leveraging an aviation-specific language model designed to identify patterns in text. This approach aims to improve data quality by detecting and presenting potential errors to a reliability engineer, who still retains ultimate decision-making authority. The result is a more efficient process that retains human oversight.

  1. Automated fault identification, troubleshooting, and repair:

When a fault occurs, the technician is often required to spend considerable time analyzing the manual for fault isolation, with the goal of accurately identifying the source of the problem and determining the steps necessary for resolution and appropriate repairs.

In addition to fault classification, the use of an aviation-specific language model could occur in real time upon detection of a fault. This model would be useful in identifying potential sources of the problem, suggesting resolution activities, and proposing remedial solutions. By presenting these options to the technician, the model could alleviate the workload and enable more efficient use of time. By including the past success rate of each option, the model would enable the technician to select options that save time and solve the problem more quickly. Resolving the failure on the first attempt can help the aircraft return to flight sooner and even prevent recurrences in the future. Avoiding or reducing delays has real value for airlines. According to Airlines for America, by 2023 delays will have a direct cost of $101.18 for every minute a flight is delayed.

  1. Predictive maintenance and anomaly detection

The concept of predictive maintenance is not new. What is presented as innovative is the application of new types of Artificial Intelligence, specifically Anomaly Detection and Pattern Recognition, which are making predictive maintenance much more accessible.
Early predictive maintenance models were limited to looking only at historical data. The introduction of IoT and real-time data feeds from sensors has made it possible to include current and up-to-date data. However, the vast amount of data generated required highly skilled data scientists. With the advent of Machine Learning, data scientists were able to focus on creating the learning model for Artificial Intelligence, freeing them from the burden of direct data processing.
Unsupervised learning models are further lowering the barrier to entry for the use of Artificial Intelligence in predictive maintenance applications. Despite the fact that the concept of “unsupervised” learning models may seem intimidating, it actually indicates the ability to integrate AI into a data set, allowing it to develop its own algorithm independently. Not only does this reduce the time and cost required to implement a solution, but it also has the power to remove bias from the process, especially when handling large amounts of unlabeled data, as is the case with the vast stream of data from the sensors of a modern aircraft or jet engine: the latest generation of aircraft will produce between five and eight terabytes of data per flight by 2026, according to predictions by consulting firm Oliver Wyman.

Anomaly detection enables the integration of artificial intelligence with data from sensors. By identifying what is considered “normal,” the system can alert when deviations from that standard state occur, serving as an early warning system. When combined with pattern recognition, artificial intelligence learns to identify specific configurations in the sensory data that indicate the likelihood of certain events. The result is an accurate and reliable early warning system capable of predicting impending events. The anomaly detection capabilities of IFS.ai and Falkonry AI offer the opportunity to democratize AI, allowing easy access and interpretation of large amounts of data generated over long periods of time. This makes, at last, predictive maintenance accessible to all.

The future is AI, the future is now.

AI should undoubtedly be employed with great care, especially in the aviation industry. The aviation industry can make significant progress in efficiency and accuracy through the use of AI to optimize human work and provide decision support. This not only helps to reduce information overabundance, but also ensures that the human maintains a central role and remains the primary decision maker in the final decision. By synergistically integrating AI with human experience, organizations can achieve optimal results and gain a competitive advantage in their industry. These improvements can generate tangible value for airlines, aircraft operators, and ultimately for customers. AI is now a reality that is here to stay, and those who embrace it first have the opportunity to gain an unprecedented advantage.

We have the support you need
We are here for you!

Solutions