Keynote Talk #1
Natural Language Processing and Artificial Intelligence as enablers for PHM Applications
Dr. Robert Plana, CTO, ASSYSTEM E&I, France
Robert Plana has been Professor at Paul Sabatier University in Toulouse and at the Institut Universitaire de France in the field of Internet of Things Technologies. He has occupied numerous top management positions at CNRS, National Research Agency and Ministry of higher education and research. In the private sector, he has been working with a startup (SiGe Microsystems) as Senior Technology Leader, with Alstom as Open Innovation Director and with GE as the CTO and Ecosystem Director of the GE Digital Services for Europe. He is now the Chief Technology Officer of ASSYSTEM Group in charge of the Innovation portfolio and leading the digital transformation program.
Keynote Talk Abstract
The Maintenance of critical infrastructures is a rather complex operation as we need to make sure that the maintenance will not translate into some downtime or will respect the very strict requirements mostly in term of safety and security. To accommodate with these issues, the maintenance operations are done through periodic inspections involving a lot of manpower and tasks that could not be necessary. Additionally, it has to be outlined that the introduction of new technologies (sensors, wireless connectivity) is not easy due to the qualification of these technologies.
In this presentation, we will firstly report on the main requirements that are mandatory to maintain two categories of mission critical infrastructure represented by nuclear power plant and railway infrastructure. It will be shown the difficulty to introduce the Internet Of Things technology and to implement a PHM strategy to optimize the maintenance of these infrastructures.
In a second part, it will be pointed out that during the design, construction and commissioning of the infrastructure, the data that are collected are very rich in term of knowledge of the health of the infrastructure. The expert’s knowledge that can be encapsulated through models and/or algorithms is also very important as it gives some indication concerning the fragility and the sensitivity of the infrastructure or certain sub-systems of it.
During the maintenance operations, a lot of documentation is provided concerning the processes that have been carried out but also concerning the incidents that have occurred.
We will show that NLP techniques coupled to Artificial Intelligence will translate to a better understanding of the potential failure modes and that it will be possible to predict the future failure. Doing that will prevent to the introduction of more sensors and connectivity and will allow to deploy a PHM strategy with a minimum of “time series” data, Most of the information will be included inside the reports and documents created during the operations of the infrastructure.
The Digital Twin approach that is going to be largely used for predictive maintenance applications will be also applicable but will involve the expert’s knowledge, Machine learning and deep learning techniques, models originating from the design phase of the infrastructure.
As a conclusion, the presentation will show that for critical infrastructure, the PHM applications will be built through a Model Based System Engineering that will create the data continuity and that will define the boundary conditions for calculating the health index of the infrastructure. NLP and Artificial Intelligence techniques will be essential to create digital twin as a system level but also at sub systems that will allow the calculation of the Remaining useful time of the assets and that will translate to define a completely disruptive approach for the maintenance. Some operations could be done during the operations of the infrastructure, others could be postponed due to a better understanding of the health of the different assets that will change the organization and the scheduling of the maintenance. This will turn into a more efficient maintenance strategy respecting the schedule and optimizing the cost of the operations.
Keynote Talk #2
Application of Oil Debris Monitoring and Analysis for PHM in Aviation and Energy
Ms. Pooja Suresh, Director of Research & Innovation, Gastops, Canada
Pooja Suresh is the Director of Research & Innovation at Gastops in Ottawa, Canada. She leads the R&D group on several programs that range from studying how rotating machinery fails, to development of technologies to detect the failure symptoms, to prognostic and diagnostic algorithms to assess the level of failure. She is currently leading a program to integrate several condition monitoring data sources for predictive maintenance through artificial intelligence techniques and hybrid algorithms that combine data-based and expert-based models. With 10 years of experience in the aviation industry, she also has a Master’s degree in Aeronautics & Astronautics from MIT.
Keynote Talk Abstract
Owners, operators, and maintainers of critical equipment need reliable and timely information on the condition of their equipment to avoid unplanned events and maximize the productivity of their assets. For industries such as aviation and energy, wear and failure of critical components in the engine and gearbox such as bearings, gears and seals are major reliability, safety and cost drivers. As these mechanical components wear, debris from these components is shed into the lubrication system. For example, the major damage modes of wind turbine gearboxes are bearing spall and gear teeth pitting, both of which release metallic debris particles in the oil lubrication system. Monitoring and analyzing this wear debris as part of condition-based maintenance programs can be effective for providing early indication and quantification of component damage, thus converting costly reactive or scheduled maintenance programs into proactive, planned and cost-efficient programs. Particle characteristics such as size, shape and composition can be used to determine the modes, sources and stages of wear in the equipment. Identification of the alloy composition of the individual particles and correlation to size and number of particles obtained of the specific alloy can directly pinpoint the damaged component and thereby, the criticality of the debris release event. Advanced oil debris sensors and analyzers combined with an expert system and predictive analytics are well suited for prognostics and health management (PHM), and provide the ability to detect damage initiation and progression, assess the damage severity, predict remaining useful life and improve maintenance actions.
This presentation reviews the application of oil debris monitoring and analysis as an effective PHM solution in the aviation and energy industries. This presentation will first describe the technologies and concepts of operation for oil debris monitoring and analysis, and then explain the typical bearing/gear failure modes and their characterization based on the generated wear debris. This is shown through accelerated failure test data and field application data. The presentation outlines physical damage severity models as well as the potential for data-based machine learning approaches for prognostics. Finally, the application of oil debris monitoring and analysis as an effective PHM solution is illustrated by presenting case studies of field events.