Members of our team worked with an Aerospace manufacturer to help them apply advanced machine learning to whole aircraft predictive maintenance. In this project the customer desired prognostic capabilities developed for components that had less instrumentation and therefore data than is available for the most critical systems in the airframe.
Our approach to this problem designed a system in which the customers engineers could work together with data scientists to follow an engineering led process to create useful features for machine learning. Key aspects of this system included the ability to work with petabytes of high frequency data in which 90%+ contained data that did not exhibit symptoms of poor health.
By approaching this problem from the fundamentals of machine learning and the engineering principles of why the machines failed we were able to build a useful predictive maintenance solution for the customer.
Trust in machine learning in this domain needs to be built up by contextualization. We realized early on that a simple email to alert of potential failure would not get the support of teams responsible for high value, high integrity assets. To counter this problem we pioneered a decision support tool approach that contextualizes the predictions. Being able to show historic outcomes based on similar predictions allowed the engineers to buy into the system and start trusting its output and also feel empowered to work with our data team to improve input features where required.
Please contact us to find out how we can help you with bespoke predictive maintenance solutions.