NON-PEER REVIEW
20
th
Australian International Aerospace Congress, 27-28 February 2023, Melbourne
20th Australian International Aerospace Congress
ISBN number: 978-1-925627-66-4
Data-driven Prognostics and Diagnostics for Turbofan
Engine
Emmanuel Blanchard
a
, Peter D M Brady
a
, Russell Graves
b
, Peeyush Pankaj
c
, Rachel Johnson
b
, Vineet J Kuruvilla
d
a
MathWorks Australia
, Level 6, Tower 2, 475 Victoria Avenue, Chatswood, NSW, Australia 2067
b
MathWorks USA, 3 Apple Hill Drive, Natick, MA, USA 01760-2098
c
MathWorks India, Trillium Building, Blocks I & J, Embassy Tech Village, Bangalore, India 560103
d
MathWorks Singapore, 10C, #06-49, Ubi Techpark, Singapore 408564
Abstract
Remaining Useful Life (RUL) of a machine is the expected life or usage time remaining before
the machine requires repair or replacement. Reliable RUL estimation can bring many benefits
to OEMs and machine operators, such as cost saving through optimised maintenance
scheduling, longer machine uptime, and reduced unexpected downtime. It also opens the
possibility of creating a new revenue stream by providing Predictive Maintenance as a service.
In this work, we use the N-CMAPSS dataset to describe a workflow and solution to achieve
two goals: (1) detect and classify faults in a turbofan engine; (2) estimate the RUL once we
detect performance degradation.
We analyse, pre-process and extract/engineer key features from the multivariate time series raw
sensor data by leveraging our understanding of how gas turbines operate (e.g., Brayton Cycle).
We also analyse the performance of various engine submodules for different flight phases
(climb, cruise, and descent). We train and compare multiple machine learning models before
using a neural network model to differentiate between healthy operation and seven different
types of faults in the turbofan engine. We train an exponential degradation model for RUL
prediction after evaluating the features' monotonicity, trendability, and prognosability. In
addition to fault detection and classification and RUL prediction, we also describe an approach
to downsample the time series data without losing information relevant to our goals.
Keywords: Predictive maintenance, remaining useful life, fault classification, N-CMAPSS,
RUL
Introduction
Predictive maintenance can be considered the holy grail of industrial machinery equipment
manufacturers and operators. It helps monitor the health of equipment to estimate its Remaining
Useful Life (RUL). These techniques will help transition from reactive maintenance to a
preventive and optimised maintenance strategy. There is immense value to gain from having a
proactive maintenance strategy, such as cost savings [1], productivity increase for the
maintenance crew, and even opening new service/revenue streams [2].
Various approaches are used for developing predictive maintenance techniques. We can broadly
classify them as model-based methods and data-driven methods. This paper focuses on a data-
driven approach to aircraft engine prognostics and diagnostics. We used the N-CMAPSS dataset
[3] to demonstrate a predictive maintenance development workflow, and we answer the three
main questions for any predictive maintenance application:
1. Is our aircraft engine or engine components’ health degrading at an abnormal rate?
2. Which subsystem(s) is failing?
3. How many flight cycles remain before the engine fails?
