Due to the Covid-19 sitation, the conference is moving online. This is an exiting opportunity to secure the participation of internationally recognised speakers from the US and Autralia among others! Also, we are preparing a nice program with many networking opportunities!
The conference will take place on the 08-th and 09-th of September 2020 over two tracks: railway and other industries. The program is designed such that you will have the opportunity to attend all presentations.
The SMC becomes the IMC:
The Smart Maintenance Conference has been growing for the last three years. It evolved toward a bigger event, with more speakers, more tracks, more attendance from many more countries!
Strong of this success, we believe it is time to anchor the conference internationally and to make a clear distinction with other similar events organised throughout the world. This led us to rename the conference as the INTELLIGENT MAINTENANCE CONFERENCE. This change of name does not change our objectives: to be place to share fruitful discussions on predictive maintenance between experts from the industry and from the academic world.
In the same spirit as in the past events, we are preparing a conference with renowned speakers from varied backgrounds, including railway, automotive and aerospace industry, academics and others. The program is already available, you should have a look!
|Prof. Olga Fink
|Session Chair: Dr. Kai Hencken
|Prof. Melinda Hodkiewicz
Dr. Débora Corrêa
|University of Western Australia
|Session Chair: Prof. Enrique Lopez Droguett
|Dr. Wan-Jui Lee
|Dr. Oliver Cassebaum
Andreas Udo Sass
|Digital Ecosystem, Volkswagen AG
|Session Chair: Shalini Trefzer
|Dr. Pierre Dersin
|Alstom Digital Mobility
|Prof. Diego Galar
|Panel Chair: Prof. Olga Fink
| Panel: From idea to implementation: challenges and lessons learnt along the way
Jan Mys (InfraBel), Dr. Pierre Dersin (Alstom), Dr. Matthias Graeber(Bühler)
|Session Chair: Prof. David W. Coit
|Dr. Matthias Graeber
|Dr. Abhinav Saxena
|GE Global Research
|Keynote Chair: Prof. Olga Fink
|Dr. Chetan Kulkarni
|Prof. Olga Fink
Dr. Oliver Cassebaum and Andreas Udo Sass, Digital Ecosystem, Volkswagen AG.
In the automotive industry, a large amount of test data is generated along the entire development process. These data are currently analyzed only to a limited extent. CAN-logs from real life endurance testing, having often more than 10^9 samples, contain valuable information regarding the robustness and aging behavior of components.
One challenge is the definition of the needed dataset for series applications in order to minimize the data stream and to select a valid subset of features in order to generate a suitable accuracy of component ageing estimation, also for unknown vehicles. In the first part of our presentation we show one way how to deal with that.
Another challenge in analyzing is to find reproducible and therefore exactly comparable load situations. Additionally, current diagnostic methods are designed to detect well known errors but are not capable to identify any unforeseen anomalies. A greater prospect of success to uncover unknown component changes offers the comparison of similar load situations. In this presentation novel machine learning methods are presented, which identify such typical similar load situations and detect them in the data set.
José Martínez, Solenix.
In this talk we will discuss which challenges are being tackled with Machine Learning in the Space Operations Domain. The topics include anomaly detection, anomaly investigation and prediction.
Anomaly Detection: our approach is currently being used by the European Space Agency to perform early detection of potential problems. We introduce the novelty detection philosophy we have implemented, how we minimise the number of false alarms and presents the results to Flight Control Team engineers. In addition, we will discuss current research in contextualizing anomalies and our attempt to build a Space Operations Dataset for anomaly detection benchmarking.
Anomaly Investigation: in case there is an anomaly, space operators need to understand what caused it to prevent the anomaly to happen again in the future. We use simple data analytics to reduce the information overhead and point to flight control engineers which are the most promising directions for further investigation. In addition, we try to build a dependency graph from data alone that may help in developing a further understanding of the interdependencies among telemetry parameters.
Prediction: we use machine learning to forecast thermal power consumption, collision risk between satellites and space debris or when satellites will cross the radiation belts. Sometimes, we also use machine learning for more Earthy endeavours such as predicting which ESA news will be more popular in front of an international audience.
This talk will give you an overview of these different applications of space operations and will provide you with references in case you would like to go deeper in any of the topics.
Dr. Abhinav Saxena, GE Global Research.
Predictive Maintenance (PM) is becoming ubiquitous for improving availability and reliability along with reducing O&M costs in industrial systems. Despite significant research and development investment in the last decade most deployed solutions still tend to be piecemeal (component or failure mode specific) point solutions and generally lack trust with respect to automated decision making. Full end-to-end deployment with system-wide coverage and autonomy still remains an elusive goal in industrial setting. This is primarily due to high cost and limited scalability of conventional modeling approaches for underlying complex systems and processes in large fleets. Specifically, capabilities to safe-guard against unknown-unknowns, lack of explainability and trust tend to be key bottlenecks. Given these systems are heavily instrumented generating large volumes of high-speed data and compute costs continue to go down, recent advancements in data-driven methods using machine learning (ML) and artificial intelligence (AI) have shown promise in a number of areas that previously led to valley of death between PM technology and commercialization.
GE’s Digital Twin technology for Predictive Maintenance is leveraging AI to bridge a number of such critical gaps that were otherwise very challenging to tackle through conventional methods. This session will enumerate key challenges in enabling system-wide predictive maintenance and how AI is being used to overcome these. Specifically, a causal deep learning-based approach will be described that provides a causal graph of inter-variable relationships allowing validation of deep learning model with domain experts. Further, by providing causal factors for identified anomalies root cause analysis can be facilitated for alert disposition in efficient manner at the fleet level. We will also describe our approach towards competency awareness of AI models, which aims to solve uncertainty management and trust for industrial applications of AI. Various applications and use-cases will be shared to show effectiveness of AI and ML using both structured and unstructured data in the context of intelligent PM.
Agnes Fritsch, Altran.
During the event, we will present a framework that can be used for various industrial use cases.
Different application sensors and different types of algorithms (deep learning, analytical tools) connect to this framework, which collects data and is able to offer geolocation/geofencing, it displays alerts and offers dashboards designed for the users.
We have learned from our clients that it is often important to start small and simple and then grow in complexity: it helps the operators understand how they generate savings, increase efficiency and create value as they scale up these solutions.
We will also detail a use case where the maintenance is in sensitive/dangerous zones so that any intervention required the stopping of the whole production: thus it becomes essential to anticipate failures and mutualize preventive maintenance.
To facilitate and solve the anticipation problems in preventive and predictive maintenance, we have also assessed the importance of collecting information about the robustness of the industrial equipment; also we base the first operational assumptions on the knowledge of the field operators so that we can find where to build in the additional sensors as needed.
Prof. Melinda Hodkiewicz, The University of Western Australia, Perth.
The ability to wirelessly stream data from sensors on mobile equipment provides opportunities to assess asset condition proactively.
Our streaming data is drawn from a mining industry case study, containing 23M rows (1.8GB) for a single excavator over nine months. This data has 58 numerical sensor variables and 40 binary indicators describing the conditions and status of different subsystems. In addition, data are available from the fleet management and maintenance work order systems. We focus on the hydraulic subsystem, which has 21 potential failure events reported in the period of the data.
There are signiffcant issues with the data due to the large volume, inconsistent and asynchronous recording from different sensors, 57% of rows have missing data, and uncertainties in the ground truth for the dependent variable (hydraulic subsystem failure). We demonstrate how the application of an OHLC (Open, High, Low, Close), commonly used in financial analysis, can be used to compress and manage the data. Secondly, we create a data frame of OHLC sensor data, fleet management and maintenance work order data and demonstrate the application of LASSO penalized logistic regression model for variable (sensor/alarm) selection.
We found that the variables selected by the data-driven method have similarities when compared to the selection made by experts (asset manager).
Dr. Chetan Kulkarni, NASA Ames
To facilitate and solve the prediction problem, awareness of the current state and health of the system is key, since it is necessary to perform condition-based system health predictions. To accurately predict the future state of any system, it is required to possess knowledge of its current health state and future operational conditional.
In case of next generation electric aircrafts, computing remaining flying time is safety-critical, since an aircraft that runs out of power (battery charge) while in the air will eventually lose control leading to catastrophe. In order to tackle and solve the prediction problem, it is essential to have awareness of the current health state of the system, especially since it is necessary to perform condition-based predictions. To be able to predict the future state of the system, it is also required to possess knowledge of the current and future operational conditions and flight profiles for accurate estimation of end-of-discharge (EOD) for the batteries.
Similar framework can be implemented to other complex systems and subsystems. Our research approach is to develop a system level health monitoring safety indicator which runs estimation and prediction algorithms to estimate remaining useful life predictions at system, subsystem as well as component levels.
Given models of the current and future system behavior, a general approach of model-based prognostics is discussed as a solution to the prediction problem and further for decision making. Data driven prognostics approaches have been equally used with good results in the past, where respective approaches have their own challenges to tackle. This limits their applicability to complex real-world domains: (a) high complexity or incompleteness of physics-based models and (b) limited representativeness of the training dataset for data-driven models. With the advent of internet of things for data collection and increased use of ML algorithms, hybrid approaches are the next avenue to reduce the challenges and achieve better results.
An hybrid framework for fusing information from physics-based performance models along with deep learning algorithms for prognostics of complex safety critical systems is presented. In this framework, we use physics-based performance models to infer unobservable model parameters related to the system's components health solving a calibration problem.
Dr. Wan-Jui Lee, Dutch Railways.
Due to the dynamic and complex nature of corrective and predictive maintenance tasks, it is challenging to react to maintenance requests effectively within a short time. In operations research, scheduling and planning problems are mostly solved by mathematical modelling if the (simplified) problem can be properly formulated, or by heuristic search if a locally optimal solution is acceptable. However, neither of these approaches takes into account previous solutions to similar problem instances. That is, given any problem instance, solutions need to be searched and found from scratch. Deep reinforcement learning provides potential to adjust to new problems quickly by utilising experience and knowledge gained from solving old problems.
In this talk, the application of deep reinforcement learning in solving the Train Unit Shunting Problem (TUSP) of Dutch Railways will be introduced. TUSP is such a complex problem that it often costs a human planner an entire work day to figure out a 24-hour maintenance schedule for a specific shunting yard. Its complexity also makes it difficult to have a proper optimisation formulation; most attempts in the literature either simplifies the problem or only focus on sub-problems. By exploring the most challenging problem in transportation, we would like to identify the feasibility, advantages and potential issues of deep reinforcement learning towards its integration and implementation within the maintenance process.
Mathias Pawlowsky, Axpo
Axpo is Switzerland's largest producer of renewable energy and an international leader in energy trading. A large share of the produced renewable energy stems from the about 60 hydropower plants operated by Axpo in Switzerland.
To further improve the safety, reliability, and efficiency of these power plants, Axpo has initiated the Hydro 4.0 project. Hydro 4.0 investigates the potential benefits of digitalization for the operation of hydro power plants in 20 use cases.
Part of Hydro 4.0 is to use data science to better understand the condition of hydro power plants. To this end, Axpo has started collecting data from three power plants by connecting their control systems to the cloud (this corresponds to recording 50’000 signals, generating ~20 million rows every day). Based on a better knowledge of the condition of a power plant, Axpo can improve maintenance, asset management, and plant operation. Axpo's architecture for data ingestion and processing from geographically spread power plants, as well as a selection of machine learning models built upon the power plant data, will be presented.
Prof. Diego Galar, Lulea Univesity
Transportation and railway in particular is suffering revolution in which artificial intelligence applications, from virtual assistants to advanced robotics, disrupt end-to-end value chains amid radical shifts in demand. The scope of change compels many manufacturers to adopt new plant designs, reshape their manufacturing footprints, and devise new supply chain models. Advances in AI technologies is enabling railways to leverage rapid growth in the volume of data to optimize processes in real time.
The talk will show how the growth of industrial artificial intelligence (AI) is reimagining the transportation sector in many dimensions. Companies are learning how to use their data to not only analyze the past but predict the future as well. Indeed, one industry that can expect to see unprecedented savings from AI is railway where AI can usher in a new era. AI is not just helping with failure predictions; it is also supporting operators on the front line to understand their assets even better than before.
However, the great benefit of the AI in transportation is the accurate prediction of the performance and the failure in order to replace the traditional O&M. While effective, there are certain drawbacks to these methods since robustness and resilience of AI are still unclear in some sectors. In fact, it’s not an exact science and risks must be properly quantified or else AI triggers some fears for the deployment in the field. This disruptive transformation of railways is traumatic that is why confidence in the industrial AI is crucial.
That confidence is not easy to gain and data quantity and quality are essential. Much information needs to be captured and mined to assess the overall condition of the whole system including the one from design and manufacturing which obviously contains the physical knowledge. Therefore, the integration of asset information during the entire lifecycle is required to get an accurate health assessment of the whole system avoiding the catastrophic failures known as black swans. Indeed, the lack of data on advanced degraded states due to early replacements and “black swans” makes the data-driven approach vulnerable to such situations. The risk related to these scenarios, despite their low latency, is not acceptable, especially for this type of assets for which safety is a must. Therefore, there is a need to augment datasets before training data-driven algorithms. For this purpose, Data covering a wider range of scenarios can be obtained by synthetic data generated by physics-based models. These models need to be realistic and provide meaningful and comparable information about the behavior of the system under observation.
Industrial AI will help the use/owner/maintainer/designer to perform a virtual commissioning of the asset where it is digitized and virtualized combining the existing physical models with the data collected from the field and produce a digital twin containing both data driven and physical information. This virtualization by Industrial AI means allows the user to produce data regarding situations and scenarios which didn’t happen yet or are very rare. These new data sets can be blindly fused to obtain a hybrid model and go one step beyond the digital twins.
Michel Kunz, SBB.
The success of the Swiss Railway System comes at a cost. It is not only necessary to increase the efficiency of ongoing maintenance tasks, we also have to take more effective maintenance measures in the future. SBB has therefore launched multiple, innovative maintenance initiatives.
Technical innovations hold without any doubt tremendous value enhancement potential. These innovations will pave the way for advanced predictive maintenance solutions, which are based on artificial intelligence algorithms.
From a management’s perspective however it must be mentioned critically, that it will not be enough to focus on purely technical solutions and isolated problem solving. We all know, that more data by itself, will not necessarily lead to better decision making but will, if not dealt with properly, create more complexity. Therefore, new forms of collaboration and cooperation will be required to ensure that maintenance is not just «smart» but rather «clever».
Part of the solution will be platforms, which not only gather available data and deliver analytical tools but also foster cross disciplinary cooperation and learning processes. Two attempts to create such clever platforms that incorporate necessary data, analytics and data cooperation amongst users will be presented in this talk.
Dr. Pierre Dersin, Alstom Digital Mobility.
What is smart mobility ?
Smart Mobility is the adaptation of transportation supply to the smart city of today, a combination of sustainable transport solutions that meet in quasi-real-time the needs of the 21st century city dweller or suburbanite. Smart Mobility systems need to be multi-modal, proactive, reactive and resilient. They include a panoply of diverse offers of which rail constitutes the backbone but is complemented by road modes such as busses and also autonomous vehicles, cycles, etc. Together they constitute a system of systems whose overall goal is to carry passengers safely, punctually and seamlessly.
Why does smart mobility require “smart maintenance” ?
Maintenance as traditionally performed consists of scheduled (time-based or distance-based) preventive maintenance and corrective ( unscheduled) maintenance. This traditional solution is wasteful of resources and its inherently rigid character is not suited to the objectives just outlined. At least some measure of condition-based or predictive maintenance is therefore in order in the smart mobility context. But only by considering the entire fleet and through a holistic approach encompassing both maintenance and operation can such an approach meet the needs of the smart city. These considerations lead to the concepts of dynamic maintenance planning and predictive traffic management.
Alstom’s achievements in those areas will be presented, such as the Fleet Support Centers, the Demand Optimizer and the traffic management simulators. Perspectives and corresponding challenges include:
Qin Wang, ETH Zürich.
Thanks to digitization of industrial assets in fleets, the ambitious goal of transferring fault diagnosis models from one machine to the other has raised great interest. Solving these domain adaptive transfer learning tasks has the potential to save large efforts on manually labeling data and modifying models for new machines in the same fleet. Although data-driven methods have shown great potential in fault diagnosis applications, their ability to generalize on new machines and new working conditions are limited because of their tendency to overfit to the training set in reality. One promising solution to this problem is to use domain adaptation techniques.
Inspired by its successful implementation in computer vision, in this talk, we will introduce several domain adaptation to our fault diagnosis context. We then carefully justify the applicability of these methods in realistic fault diagnosis settings, and offer a unified experimental protocol for a fair comparison between domain adaptation methods for fault diagnosis problems. In addition, we will also show a common challenge we face when adopting domain adversarial methods for PHM applications. We will demonstrate that the performance of domain adversarial methods can be vulnerable to an incomplete target label space during training. To overcome this issue, we propose a two-stage unilateral alignment approach. The proposed methodology makes use of the inter-class relationships of the source domain and aligns unilaterally the target to the source domain. The benefits of the proposed methodology are first evaluated on the MNIST→MNIST-M adaptation task. The proposed methodology is also evaluated on a fault diagnosis task, where the problem of missing fault types in the target training dataset is common in practice.
Salomé Iglesia, Siemens Mobility.
Currently It is possible to resolve 70% to 80% of maintenance requirements based on diagnostic data. In the speech we will talk about how to improve the remaining 30%-20%. We will answer the following questions and explain our status quo and methodology:
In summary, Siemens Intelligent Maintenance Solutions focus on minimizing disruptions of the rail operation due to corrective maintenance activities.
Dr. Matthias Graeber, Bühler.
The Swiss and family-owned company Bühler Group is a global solution provider for the food and advanced materials industry. Every day billions of people come into contact with products manufactured on Bühler machines and lines. With this global relevance also comes responsibility to optimize processes and enhance sustainability in plants and along global value chains. Bühler has partnered up with the Swiss Data Science Center to further optimize manufacturing processes using data, with respect to uptime, specific energy consumption and quality. For example, Bühler has recently launched a new digital service which allows operators to “replay” all past events in the plant control system. The corresponding data represents an accurate digital twin of the manufacturing process. In this presentation we demonstrate how this ‘process’ digital twin data can be complemented with data from maintenance systems towards adaptive plant maintenance schedules.
Vita: A physicist by education Matthias has over 10 years experience driving innovation in different industries. Currently he works as Head of Data Science at Bühler, responsible for building an in-house data science capability and partner network from ground up. Bühler’s digital teams are exploring and pioneering new digital services based on IoT data with value for the industry. Examples include data-driven yield and uptime optimization or mobile-phone based solutions for quality control.
Jan Mys, InfraBel.
Infrabel updated its strategy in 2019 with the introduction of its plan GO! With this plan Infrabel wants to put its clients even more in the center of its focus. Through operation excellence it wants to improve its performance as infrastructure manager.
Digitalization is one of the main leavers for infrabel in its search for operation excellence. Since many years digitalisation was impliemented to improve its proceses. On different subjects in asset management Infrabel has delivered beautifull successes but the process takes time and the scope in which digitalization can help is wide.
In the presentation an overview will be given of the approach Infrabel follows in order to bring digitalisation to succes in different domains of asset management. A large focus was put to improve maintenance of tracks, over catenary to signalling were improvements can be noticed at safety, reliability and efficiency. But the range of potential improvements are much broader. Digitalization is now seen as a mayor driver to improve the safety of our track workers.
During the presentation some basic criteria will be highlited that are crucial for succes by different exemples that have been implemented in the past are under roll out.
Agnes Fritsch, Altran.
Head of the Chair of Intelligent Maintenance Systems,
Chair of Intelligent Maintenance Systems,
Chair of Intelligent Maintenance Systems,
Dipl. Kulturing. ETH Zürich,
Chair of Intelligent Maintenance Systems,
Prof. Olga Fink, ETH Zürich
Dr. Gabriel Michau, ETH Zürich
Patricia Marty, ZHAW