Let’s face it, automation is everywhere.

It’s at conferences, in our pockets, in our homes, and yes, increasingly in our cockpits too. When implemented effectively, automation can save lives and enhance performance. There is a significant difference between 1,000 and 10,000 litres of fuel, and automation in Electronic Technical Logbooks (ETL/ELBs) can prevent errors by improving data accuracy and consistency.

In eTechLog8, we provide users with warnings or even hard stops when data appears incorrect. For example, we won’t allow the user to enter a fuel uplift significantly greater than the tank’s total capacity, and if the fuel uplift maths doesn’t add up, we provide warnings. These kinds of safeguards help prevent simple, costly mistakes.

Here’s the catch: not all automation is created equal.

In aviation, mistakes can lead to serious legal and safety implications.

Let’s not forget: ETL/ELBs are legal records, which means that every entry must be accurate, timely, and above all, trustworthy. Unfortunately, some of the latest kinds of automation like Generative AI can be unreliable. These systems have been known to confidently produce incorrect information, fabricate content entirely, or worse, as New York City discovered, suggest illegal actions in the name of efficiency!

Just because we can automate something doesn’t always mean we should. As James Russell, Development Director at Conduce, succinctly put it:

“Artificial Intelligence is a useful tool. However, how we use it must be governed by whether it is suitable for the task and whether it is safe to use.”

What is AI?

When people speak of Artificial Intelligence, or AI, most people aren’t exactly sure what is happening on a technical level. That’s why it’s often called a “black box”: you can see what goes in and what comes out, but the process in between can be hard to understand, even for experts. The term ‘AI’ implies some form of true intelligence. We imagine a computer thinking for itself to solve problems faster and better than a human, like something from Star Trek or Terminator.

In fact, Artificial Intelligence is a broad concept: it's about making machines behave in a way that seems intelligent - by solving problems, understanding language, or recognising patterns.

In reality, most of what we call AI today is actually something more specific: Machine Learning (ML).

Rather than being truly “intelligent,” Machine Learning is about training systems to recognise patterns in large amounts of data and make predictions based on that. Instead of programming the exact steps, we give the system examples - and it learns from them.

Machine Learning is really good at assisting with multi-class classification. Multi-class classification just means sorting things into more than two labelled buckets, like picking whether a fruit is an apple, banana, or orange. We already do this in aviation in lots of places, for example, when we record an ATA chapter against a Defect.

However, it’s important to remember that any AI or Machine Learning model’s predictions are only as good as the data on which it was trained! More about that later…

So How Can AI Help eTechLog8?

At Conduce, we are committed to remaining at the forefront of latest developments, and if they are useful, bringing them into eTechLog8. But we have to be careful – like James said, we have to look at the suitability and usefulness, and assess any risks well before we begin developing!

We recently identified somewhere within eTechLog8 where we thought Machine Learning could be suitable and useful, and didn’t bring much risk.

We realised if you feed a Machine Learning model thousands of past defect descriptions and their correct ATA chapters, it can start to predict the right chapter for a new defect by spotting similarities. It’s not thinking like a human - it’s identifying patterns and applying probabilities.

For anyone who doesn’t know, ATA chapters are like a big reference book where each numbered chapter tells you which part of the plane has a problem - for example, engines, brakes, or lights.

ATA chapters are helpful because they make it fast and easy for everyone to understand exactly which part of the plane is being talked about, no matter the airline or manufacturer. In addition, we can group and analyse data by chapter. This makes it easier to spot recurring issues, track reliability trends, identify problem areas and report on the data.

When a defect is logged in eTechLog8, there are around 100 possible ATA chapters it could be associated with.

Engineers are experts - many even know the chapter codes by heart - but scrolling through hundreds of options is time-consuming.

This is where Machine Learning can add value. It can analyse the defect description, compare it with historical defect descriptions and associated ATA codes, and suggest the chapter that it is most confident matches the description.

However, Machine Learning should only suggest the most probable ATA chapter. Even predictions with 99% confidence require human oversight. Ultimately, final decisions must always rest with a qualified and certified human who can assess context, apply judgment, and ensure full compliance with safety and regulatory standards.

Where Conduce Stands Apart

We said earlier that any AI or Machine Learning model’s predictions are only as good as the data on which it was trained. This is an area where Conduce is fantastically positioned with eTechLog8!

eTechLog8 benefits from a large volume of high-quality, clean data related to ATA chapter classification. For each defect, we record the defect description [input], and the expert’s judgement – in the form of an assigned ATA chapter [output].

This direct relationship between input text and output ATA Chapter classification allows for effective training of the Machine Learning model and measurable accuracy.

Lead Developer of the project, Joseph Sleiman, explains:

“eTechLog8 provides a high-quality and clean data set that is rarely seen, let alone available, in one place. On top of that, we are drawing on engineers’ expertise to train the AI model itself. The model couldn’t ask for better teachers.”

AI Tailored to Each Customer

Conduce takes things further by training each AI model using a customer-specific Machine Learning Ops process.

Defect data is being recorded every day, on every aircraft, by engineers using eTechLog8. This information is collected and combined using an automated pipeline, cleaned to remove human errors, and then formatted into a CSV file that is used to train the AI model.

The Machine Learning process uses 80% of the data to teach the model. The remaining 20% of data is used to validate and test the model’s accuracy.

We can keep training the model and running the tests until we have a final version, with the highest accuracy – it suggests the correct ATA Chapter most of the time. This is then uploaded into the customer eTechLog8 system. This results in a highly tailored, accurate, and efficient AI model designed specifically for that customer, in turn, reducing the risks associated with generic AI systems.

This also means we can continuously improve the model. Once it’s in use, we can compare the predicted chapter to what the engineer actually selected, expanding our clean, reliable and tailored dataset. Or in Joseph’s words:

“It’s of the best facets of the model. We can take into account discrepancies between the engineer and the model, and continuously improve the model by repeating the same cycle/pipeline, eventually deploying more and more accurate models to the customer, as they record more data!”

So How Is It Implemented in eTechLog8?

When completing defect actions, our customer-specific Machine Learning model displays a bold, starred suggestion at the top of the ATA chapter list picker.

This recommendation can always be overridden by the engineer’s expertise, but displaying the most likely option at the top of the list, rather than having to scroll through to find the right one saves valuable time during the aircraft’s turnaround. It also improves accuracy- not because engineers don’t know the correct chapter, but because it reduces mistakes caused by simple “finger trouble”, those accidental mis-taps on the screen.

Conclusion

So there you have it – a small, safe, but very powerful way that Conduce are implementing AI and Machine Learning within eTechLog8. Several customers are already using this feature, and seeing the model correctly predict the ATA Chapter over 86% of the time – and that’s only based on the first set of training – we should see the system get more and more accurate over time.

We’ll keep on looking at places we can use AI to make Engineers and Pilots lives easier - enhancing efficiency through eTechLog8, without ever compromising the accuracy, accountability, or integrity of the Technical Logbook. Watch this space for further developments!

If you're interested in this feature or starting your own eTechLog8 project, get in touch with you Conduce Project Manager, contact us here, or send us an email at info@conduce.net