Predictive maintenance

Objectives and skills acquired
Keeping industrial systems in operating condition at the lowest possible cost has become a critical factor in business performance, and traditional maintenance concepts are gradually being supplemented by a more proactive approach to failures. With this in mind, predictive maintenance is receiving increasing attention. The overall principle is to transform a set of raw data collected on the equipment being monitored into health indicators which, when extrapolated over time, can be used to define detailed reaction policies. In this module, we aim:

to present the emergence of this predictive maintenance theme,
to explain and illustrate the underlying processes (particularly that of prognosis),
to describe the benefits that can be expected from the implementation of these solutions,
to provide some food for thought on the challenges still facing us today.

At the end of this module, students will be in a position to:

assess the appropriateness of starting a predictive maintenance project,
deploy the associated methodology and orchestrate a set of algorithms to support the stages of this process,
assess the performance of the solutions developed.

In addition, students will be made aware of the "innovative" nature of predictive maintenance (part of Industry 4.0), the obstacles to its implementation (changes in practices), the complementarity of existing information systems (IoT, MES, CMMS, ERP), and the necessary interactions with other trades (scheduling, quality, after-sales service, etc.).).

Skills currently being acquired

Mobilize a broad field of fundamental and technical sciences related to avionics and space systems, and have the analytical and synthesis skills associated with them
Have a global systems approach

Signal processing: AP5NUTDS
Probability and statistics: AP5SISPI
SDF concept

Contents

Part 1 - Introduction to predictive maintenance

The value: illustrative examples
The vector: concepts, positioning and challenges of predictive maintenance
The foundation: basic reminders (notions of risk, failure, FMDS quantities)
The first part of the module is a general introduction to predictive maintenance. It highlights the strategic urgency of taking more proactive account of failure phenomena, and describes how the stakes, prerogatives and practices of maintenance departments have evolved as a result.




Section 2 - Predictive maintenance: methodology

Structuring the approach: CBM rosette and Prognostics and Health Management
Observing the system: data acquisition and pre-processing
Modeling/analyzing degradations: detection, diagnosis and prognosis
Taking action: optimizing, deciding and capitalizing



In visual terms, predictive maintenance aims to give meaning to raw data that carries information on the evolution of equipment pathology. In this second part of the module, we identify a coherent set of processes needed to carry out this type of analysis and make the appropriate decisions.

Strand 3 - Predictive maintenance: tools

"Conventional" stochastic approaches
Signal processing and multivariate analyses
Physical models and black/grey boxes (artificial intelligence and machine learning)
Error, confidence and predictability measures



In this third strand of the module, students are introduced to a number of algorithmic techniques for: 1) generate descriptors of equipment health, 2) model the dynamics of degradation, 3) estimate residual lifetime, 4) judge the performance of the analyses carried out.

Section 4 - Conclusions and openings

Predictive maintenance: myth or reality, data-science vs. business processes
Lifecycle, ROI and changes in practices
Current challenges



In conclusion, we take a critical look at the maturity of the predictive maintenance activity, and open the discussion on the one hand, on the "economic value" that it carries, and on the other, on the scientific and technical problems that remain largely open.
Teaching method

This module is taught in the conventional way: on the basis of introductory lectures and practical exercises.
To illustrate the elements introduced, numerous simulations and videos are shown to students.
Students are asked (in pairs) to work on a small predictive maintenance example (Matlab or Python). A small amount of bibliographical work is also required: production of a tool sheet on a predictive maintenance algorithm.

Maintenance of the Future