Advancing the possibilities of data analytics for more efficient maintenance management.
By Darren Macer, Associate Technical Fellow, Boeing Global Services
Through the use of data analytics, powerful and fast computing hardware, and ubiquitous high-bandwidth data communications, the role of software is progressing from servant to co-worker—an equal member of the team as a result of artificial intelligence, machine learning and other data science techniques. This enables the world of predictive and prescriptive analytics—the ability to predict that something will happen and then prescribe what action should be taken to either prevent or correct it.
The accuracy needed to achieve the full potential of predictive and prescriptive aerospace analytics, however, occurs from a close collaboration between data science and aerospace experts in converting information into new insights, opportunities and applications. This is the basis of Boeing AnalytX, the collective efforts of thousands of Boeing experts working together to advance aerospace and deliver a new generation of analytics-enabled products and services.
Data analytics are typically grouped into four types: descriptive, diagnostic, predictive and prescriptive.
Descriptive analytics identifies that something occurred, for example, a high temperature limit being reached. Diagnostic analytics focuses on supporting root-cause analysis.
With advances in data science, processing techniques and power; a good supply of data; and deep domain knowledge, we are able to determine the health of a component or system, and predict when it is degrading and when it will fail. This provides the basis for predictive analytics— the capability to peek over the time horizon, getting a heads-up to a potential future situation.
Armed with this information, an organization can better prepare to manage a situation in advance rather than react after it happens. Boeing embeds fault logic into our aircraft and has a long history of both onboard and off-board processing of data, through mathematical and statistical methods, to provide predictive alerts and recommendations via Airplane Health Management. Now through advances in computing power linked with data science techniques and proprietary capabilities, Boeing is able to predict the future for issues not generally detectable by humans.
What happens when an operations center receives a predictive alert? Traditionally, this is when people take over, and the software goes back to monitoring, or other software is used to support identifying and evaluating viable options.
By harnessing the techniques from descriptive and diagnostic capabilities and linking with the ability to predict an event, we are now able to know what will happen, when it will happen, and also what should be done to correct it—this is prescriptive analytics, answering “What needs to be done?”
Answering that question is just the first step of what prescriptive analytics can do. The next levels raise the software to an equal co-worker—where diagnostic, predictive and prescriptive analytics work together to take autonomous action within user-defined constraints. This leaves the more complex issues to people, playing to their strengths, such as creativity, adaptability and innovation.
The new digital frontier opened by Boeing AnalytX provides autonomous and semi-autonomous systems in support of technical operations spanning the airplane, fleet and enterprise.
It takes more than expertise in data analytics to deliver aerospace analytics capabilities to the industry, however. It also requires wide and deep industry and domain knowledge. Boeing is one of the few companies to combine analytics and deep domain expertise under one roof, enabling it to efficiently transform data into meaningful and valuable insights.
Prescriptive maintenance can offer great efficiencies and savings to an operator by looking ahead, watching for precursor events, and converting unscheduled maintenance events into scheduled ones. Predictive maintenance also helps with routine maintenance such as brake wear or oil consumption, and it enables health-based condition monitoring through detecting component or system degradation.
As problems become more challenging and signatures of degradation more subtle, the traditional engineering and physics-based approaches require assistance.
This is driving engineers and data scientists to use artificial intelligence techniques, such as machine learning. When these capabilities are paired with Boeing’s experience and expertise, early prediction and accurate prescription become possible.
We now have the capability not only to predict a failure, with sufficient time to allow for consideration and planning, but to also prescribe exactly what actions should be taken to correct the problem in a clear and methodical way.
Both prediction and prescription must be completed with the highest level of accuracy to provide a measurably better outcome than existing reactive methods or based only on the technician’s know-how.
To best illustrate the approach and capability, the following is an example of what happens when a Boeing monitored 787 aircraft lands at an airport capable of offloading data. The aircraft automatically transfers the full flight data to a ground network to be processed through the cycle detailed below, progressing to prescriptive instructions via Boeing applications.
- LSAP Librarian Suite—a unified capability for airlines to efficiently transfer large amounts of data to and from airplanes. It offloads the full flight sensor data from an aircraft and delivers it to the Boeing environment.
- MAESTRO, which stands for mass-data analytics engineering system—translator, repository and orchestrator, is an automated system that takes the binary data from the onboard recording systems and converts, translates, analyzes, stores and publishes flight recorder data in reusable engineering units.
- Boeing AnalytX Data Warehouse—Big data storage and processing capability.
- Automated assessment by a proprietary Parameter Manipulation Language easily enables research, understanding and processing of the data.
- Automated Predictive Environment—Rules engine that runs algorithms on the data and appends instructions enables machine learning and engineering-based predictive alerts to be applied to the inbound, processed data.
- Prescriptive detailed work instructions, identified through engineering knowledge and machine learning methods, are automatically appended to the alerts.
- Prescriptions, such as parts and a mechanic, are scheduled at a future stop in order to perform maintenance without disrupting operations.
This capability, enabled by deep domain knowledge and empowered by artificial intelligence benefits customers by reducing unscheduled maintenance, resulting in higher schedule reliability.
Predictive and prescriptive analytics provide views over the time horizon, extend situational awareness, and gain more time to think, plan, prepare and manage situations rather than just react to them.