...
Program 2022 Autumn
| When | Who | What |
|---|---|---|
| 15.12.2022 | TBD | More information will come later. |
| 01.12.2022 | TBD | More information will come later. |
| 17.11.2022 | Postdoc Tzioutzios Dimitrios | New Postdoc Self-introduction. |
| 03.11.2022 | Ph.D. Candidate Alessandro Campari | More information will come later. |
27.10.2022 15:00-16:00 (Digital Only) | Chi Ji | Title: Autonomy safety and SOTIF Abstract: The presentation will start from the introduction of key technologies in autonomous driving and challenges in autonomy safety. SOTIF (safety and intended functionality) will be then presented, including the approach of SOTIF analysis, acceptance criteria and SOTIF validation targets, followed with examples of calculation methods. About Speaker: Mrs. Ji has master’s degrees in both engineering and business administration. She has 12 years of working experience in automobile industry (including airbag, braking and autonomous driving). From 2010 to 2018, Mrs. Ji worked at Autoliv and ZF, she was in charge of the functional safety aspects of airbag and braking systems. After 8 years’ in Tier1, Mrs. Ji joined the SAIC Autonomous Driving Center, her main job was focused on L2+ autonomous driving, as well as system level safety analysis work. Last year she joined UL and become an Autonomy safety consultant. |
05.10.2022 | Ph.D. Candidate Michael Pacevicius | There will be no seminar this week. Instead, you are invited to Michael Pacevicius’s defense.
More information can be found here. |
| 22.09.2022 | Ph.D. Candidate Muhammad Gibran Alfarizi | Abstract: Bearings are essential to the reliable operation of rotating machinery in manufacturing processes. There is a rising demand for accurate bearing remaining useful life (RUL) predictions. The data-driven technique for predicting RUL of bearing has demonstrated promising prospects to facilitate intelligent prognostics. This paper proposes a new data-driven prediction framework for bearing RUL utilizing an integration of empirical mode decomposition, random forest, and Bayesian optimization. The proposed framework consists of two main phases: feature extraction and RUL prediction. The first phase of this framework focused on decomposing the empirical input signals using empirical mode decomposition into distinct frequency bands to filter out irrelevant frequencies and determine the fault characteristics of the signals. In the second phase, the RUL prediction is then carried out by an RFs-based model with its hyperparameters tuned by Bayesian optimization. The proposed approach is validated using datasets obtained from an actual run-to-failure experiment of roller bearings. The experiment results significantly improved compared to the standard data-driven and stochastic approaches. (The paper can be found here). Hydrogen can be adopted as a clean alternative to hydrocarbons fuels in the marine sector. Liquid hydrogen (LH2) is an efficient solution to transport and store hydrogen onboard of large ships. LH2 will be implemented in the maritime field in the near future. Additional safety knowledge is required since this is a new application and emerging risk might arise. Recently, a series of LH2 large-scale release tests was carried out in an outdoor facility as well as in a closed room to simulate spills during a bunkering procedure and inside the ship’s tank connection space, respectively. The extremely low boiling point of hydrogen (-253°C) can cause condensation or even solidification of oxygen and nitrogen contained in air, and thus enrich with oxygen the flammable mixture. This can represent a safety concern since it was demonstrated that a burning mixture of LH2 and solid oxygen may transition to detonation. In this study, the experimental data of an LH2 release test series recently carried out were analysed by means of an advanced machine learning approach. The aim of this study was to provide critical insights on the oxygen condensation and solidification during an LH2 accidental release. In particular, a model was developed to predict the possibility and the location of the oxygen phase change depending on the operative conditions during the bunkering operation (e.g. LH2 flowrate). The model demonstrated accurate and reliable predicting capabilities. The outcomes of the model can be exploited to select effective safety barriers such as a water deluge system to prevent the oxygen change phase. About Speaker: Muhammad Gibran Alfarizi has a B.Sc. in Petroleum Engineering from Indonesia and a M.Sc. in Petroleum Engineering from NTNU. His research interest include data driven modeling and machine learning. He started his PhD in August 2020. The project title of his Ph.D. is The Digital Transformation and Data-Driven Methods in the Reliability of Safety Systems. |
| 08.09.2022 | Postdoc Xingheng Liu | Title: Spatial-temporal interpolation for condition monitoring: an application to choke valves Abstract: Critical systems such as subsea choke valves works under a time-varying operating condition (TVOC), which makes it challenging to estimate the system’s health state since the health indicator (HI) may be recorded under different TVOC. State-space models together with Particle Filter are widely used for system identification when the functional forms of the relationship between TVOC and HI are explicitly given. However, assumptions on how the TVOC can influence the observation and the growth of HI are generally hard to validate in practice. In this presentation, we introduce the spatial-temporal interpolation methods for state estimation and prognosis for subsea choke valves. Spatio-temporal interpolation is the task of estimating the unknown values of some property at arbitrary spatial locations and times, using the known values at spatial locations and times where measurements were made. The estimated property varies with both space and time, with the assumption that the values are closer to each other with decreasing spatial and temporal distances. For subsea choke valves, the HI (flow coefficient deviation) is the quantity to be interpolated, the operating condition (percent travel) constitutes the 1-D space dimension and the timestamps at which the observations were made form the time dimension. Two popular methods, namely spatial-temporal Inverse Distance Weighting and Universal Kriging, are fitted to the data, before being compared to some other competing models (ARIMA, Wiener process…) We highlight the difference between the fundamental assumptions in these models and showcase the pros and cons of each model in terms of forecasting. Keywords: condition monitoring, state estimation, spatial-temporal interpolation, Kriging, Forecasting. |
25.08.2022 10:00-11:00 | Ph.D. Candidate Tom Ivar Pedersen | Title: Industry 4.0 and Smart Maintenance Abstract: In this RAMS seminar, I will present the results of my Ph.D. project, which is soon to be completed. My Ph.D. project belongs to the research program BRU21. BRU21 stands for Better Resource Utilization in the 21st century and is NTNU’s research and innovation program in digital and automation solutions for the Norwegian oil and gas industry. The main objectives of my Ph.D. project have been to explore how the introduction of digital solutions and concepts from Industry 4.0 to maintenance can help improve the competitiveness of this industry sector. Opponents: Postdoctoral fellow Xingheng Liu, PhD Candidate Endre Sølvsberg |
...
| When | Who | What |
|---|---|---|
- | - | No more meetings until January due to the holiday season |
09.12.2021 (Digital only) | Professor Shen Yin | Title: Robustness and Sensitivity of AI systems - Two Sides of a Coin Abstract: In recent years, artificial intelligence has made remarkable breakthrough, where a large number of AI-enabled systems have been developed and applied in manufacturing industry, medical care, cyber security, and many other fields. An interesting phenomenon lies in the different design targets of AI models – some should be robust to the abnormal changes of data while the others should be another way round. For example, to predict the remaining useful life of batteries, it is favorable to develop robust AI where the model is expected to be least affected by disturbances. By contrast, the sensitivity of the model might be critical in order to diagnosis of cyber-attacks, in which the AI models should be highly sensitive to malfunctions and malicious attack behaviors, keeping effective in case of any insignificant changes. From an engineer point of view, this talk will focus on both robustness and sensitivity of AI systems, which are regarded as two sides of a coin. The formulation of several typical demand-driven examples, the design approaches, and the corresponding performance will be introduced. A balance between sensitivity and robustness of AI is worth to be considered further in the R&D phase to cope with various demands in practice. About Speaker: Shen Yin received the B.E. degree in Automation from the Harbin Institute of Technology, Harbin, China, and the M.Sc. degree in Control and Information Systems and the PhD. (Dr.-Ing.) degree in Electrical Engineering and Information Technology from the University of Duisburg–Essen, Germany. |
| 25.11.2021 | Researcher Shenae Lee | Title: An approach to update the reliability performance of safety barriers based on operating experience (The paper is to be submitted for Loss Prevention 2022) Abstract: Hazardous events in process plants like the leakage of dangerous substances can result in severe damage, and such an event is often defined as the TOP event of a fault tree analysis (FTA) in a quantitative risk analysis. The input data for a FTA are often generic reliability data that are not necessarily catered for a plant specific analysis. Therefore, this paper presents an approach based on Bayesian network (BN) analysis with a focus on Hierarchical Bayesian analysis for handling situations where plant-specific data are sparse. The suggested approach is demonstrated by a case study of a pressure vessel. About Speaker: Shenae Lee is currently a researcher at MTP, NTNU (2020-2022). She finished her Ph.D. at RAMS group, MTP in 2020. She has B.Sc. in Nanotechnology from South Korea and M.Sc. in RAMS, NTNU. |
| 11.11.2021 | Ph.D. Candidate Tom Ivar Pedersen | Title: Data-exploration and possibilities for anomality detection and RUL-prediction Abstract: A dataset of maintenance records and sensor readings from a group of water cooled power cords collected from a plant in the process industry will be presented. Two master students at NTNU used this dataset in their master thesis earlier this year. They explored the possibilities for estimating remaining useful life (RUL) for these component. As part of his Ph.D., he plans to investigate this dataset further. The goal of this presentation is to discuss how this dataset can be used as basis for a paper. He has done some preliminary data exploration and found that there are two types of failures in the dataset. One type is a gradual degradation that can be tracked with a health indicator. A preliminary plan is to use some variant of the Wiener process as basis for RUL-prognosis. |
04.11.2021 (postponed due to time conflict) | Ph.D. Candidate Ewa Maria Laskowska, Emefon Dan | Title: Maintenance Optimization of Emergency Shutdown Valves (ESV) Abstract: The topic of the presentation is the optimization of maintenance and testing policy of Emergency Shutdown Valves (ESVs). ESVs are safety critical equipment used in oil and gas facilities. They are used only in case of a demand and remain passive due to normal operation. To ensure their safety ESVs have to be periodically (full proof tests) tested what often require production shutdown and leads to faster use of these valves. It is also possible to test valves online by so-called partial stroke tests (PST). Although this kind of test doesn’t impede the production process it is less reliable than full proof tests with regard to capacity of revealing failures. The aim of this work is to find the optimal maintenance strategy for ESVs while satisfying ESVs safety requirements. The modelling framework is based on the Markov state degradation model. In such a model, the condition of ESVs is defined by discrete states. The important part of the approach is to consider condition-based inspection policy, when the testing intervals are dependent on the ESV’s condition. Two types of valves tests are considered: partial (online) tests and full proof tests. They differ with regard to the test coverage (reliability) and cost of testing. At each inspection the decision about repair is made, so that valve’s condition can be improved to a better state, or no repair is made but there is a shift in testing frequency. Also repairs have different cost assigned depending on whether they are performed online require production shutdown. Parameters or variables considered in the model: - There are 2 variables (parameters) regarding state of the valve: state and coverage. - There are 2 types of inspections: Full Poof Tests (FPT) and Partial Tests (PT). - There are 2 decision variables with regard to condition of the valve: repair decision and inspection interval. - There are 3 kind of repairs: online repairs, repairs performed during planned shutdown and repairs requiring unplanned shutdown - There are 2 aspects to consider with regard to cost of maintenance. First the cost depends on the revealed condition of the valve. Second, the cost can be accordingly increased or reduced depending on whether it requires unplanned shutdown and is performed withing a planned shutdown - There are two cost figures related to condition monitoring: cost of FPT and cost of PT Speaker: Ph.D. candidate Ewa Maria Laskowska Title: New Ph.D. Candidate Self-introduction Emefon Dan - Self introduction.pdf Speaker: Ph.D. candidate Emefon Dan Note that the physical meeting will be at Gløshaugen Materialteknisk 3. etasje Holand. |
14.10.2021 12:00-13:00 | IEEE Reliability Society, Sweden and Norway Joint Section Chapter | There is no RAMS seminar this week. Instead, you can attend the Kick-off & Webinar of IEEE Reliability Society, Sweden and NorwayJoint Section Chapter. The agenda is as follows:
Speakers: Prof. Min Xie was awarded a scholarship to pursue his undergraduate study in Sweden in 1979. After graduating from KTH, he continued with his PhD study under the supervision of Prof Bo Bergman at Linköping University, and completed his PhD in 1987. He has been involved in IEEE Reliability Society activities since then, and was elected IEEE Fellow in 2005. Prof. Xie moved to National University of Singapore in 1991 and then to City University of Hong Kong in 2011. He has published over 300 journal papers and several books, and guides over 50 PhD students. Dr. and Adjunct Prof. Pierre Dersin studied at MIT, where he obtained first a MS in Operations Research, then a Ph.D. in Electrical Engineering under Prof. Michael Athans. Since 1990, Dr. Dersin has been with Alstom (St-Ouen, France), mainly active in Reliability Engineering and Maintenance of Railway Systems. In particular, he founded there the “RAM Center of Excellence”. He started Alstom’s Predictive Maintenance Program. He is currently Prognostics & Health Management (PHM) Director in Alstom’s Digital & Integrated Systems Division, as well as ‘ RAM Master Expert’. He is the author of many publications in Reliability, Automatic Control, and Power Systems, including four chapters the “Handbook of RAMS in Railways: Theory & Practice” (CRC Press,Taylor & Francis), 2018. Pierre Dersin is a member of the IEEE Reliability Society AdCom and leads the Technical Committee on Systems of Systems. |
30.09.2021 (Digital only) | Yuchen Jiang and Shimeng Wu from Harbin Institute of Technology | Title: Remaining useful life prediction based on machine learning approaches Abstract: Digital transformation and digital twin technologies have facilitate the connection, the interaction, and the in-depth integration of the physical space and the virtual digital space. Colossal amount of process data are available and waiting to be made use of. In light of this, the safety and reliability challenges in industrial systems are likely to usher novel solutions. In this talk, we will share our recent work on system-level RUL prediction based on machine learning approaches. The work is practical problem-driven and therefore the talk will focus on analysis of the practical problem we study, the data-related problems, the feature exploitation tools, and the ideas of how to achieve accurate RUL prediction. Finally, a short discussion will be made about the pros and cons of ML-based and traditional model-based approaches. |
16.09.2021 (Digital only) | PhD candidate Jie Liu, Wanwan Zhang | Title:New Ph.D. Candidate Self-introduction |
02.09.2021 (Digital only) | Postdoc Xingheng Liu | Title: Modeling choke valve erosion with dynamic system Abstract: Choke valve erosion is a major issue encountered in subsea oil production. When the valve is severely eroded, production needs to be slowed down or shut off to reduce the risk of major incidents such as leakage. Visual inspection is hardly possible for subsea equipment and therefore the monitoring of a choke valve installed at an X-mas tree/manifold relies usually on the monitoring of a degradation indicator, Cv (flow coefficient). The Cv indicates a valve's capability to let the fluid flow through a choke. Theoretically, the Cv increases monotonically at a given opening (valve travel/lift) with the erosion. The issue with using Cv as a degradation indicator is that the valve is operated at a non-constant opening. Consequently, plotting Cv against time will lead to a non-monotonic Cv curve. Inappropriate use of the recorded Cv sequences can therefore give misleading results about the degradation level and erosion rate. We propose to use a dynamic system to model the choke valve erosion. The hidden state evolving over time is the effective flow area (EFA) that becomes larger as the valve is eroded. The system model shows how the hidden states evolve each day and is established based on physical models (erosion response model, fluid mechanics equations) and Gamma process (for intrinsic erosion growth randomness). The observation model shows the relationship between observed Cv and the effective flow area, accounting for measurement noise. With some field data (daily allocated flow rate, sand rate, valve opening) and choke valve features (flow characteristics, Cv curve), we can estimate the model parameters and predict the erosion growth and remaining useful life of the valve. About Speaker: Xingheng Liu is a postdoctoral fellow at RAMS group, MTP, NTNU. His research topic is Prediction and optimization of remaining useful lifetime, which is a part of SUBPRO project. He completed his cotutelle Ph.D. in ROSAS (department of Operational Research, Applied Statistics and Simulation) at the University of Technology of Troyes (fr) and in RAMS at NTNU and earned his Master's and Bachelor's degree in Industrial Engineering at UTT. |
19.08.2021 | Ph.D. Candidate Endre Sølvsberg | Abstract: As part of the EU project Qu4lity, the Continental pilot “Autonomous Quality in PCB Production for Future Mobility” involves PCB (Printed Circuit Board) production in an SMD (Surface-Mount Device) line at their site in Sibiu, Romania. The main issue is faulty PCBs passing all in-line testing, and being sent out to customers, resulting in a high number of customer returns. The Pilot team has cooperated with domain expertise on-site in Romania to identify critical variables from the Continental Datalake and employed a PCA approach using the Mahalanobis Distance to identify customer returns as outliers and enabling the quality team at Sibiu to identify faulty PCBs before they are sent out to customers. Using a Docker container, the PCA algorithm has been automated with a Python script. An integrated GUI enables the quality team and other relevant people to see PCBs that are tagged by the algorithm as outliers, in addition to check individual unit ID tags. Through 5 months of PCB production, there have been 171 customer returns, and the quality team at Sibiu has validated that the algorithm was able to identify 169 of these PCBs as outliers, close to 99% of all customer returns in the 5 month period. About Speaker: Endre Sølvsberg has a BA in Economics and an MSc in Sustainable Manufacturing. He has lectured at NTNU in the Masters level course Scientific Methodology. He has worked at SINTEF Manufacturing as a project engineer and researcher, in projects involving I4.0, ZDM and Smart Maintenance. He has served as the technical lead for SINTEF in the Continental Pilot in the EU Quality project. The working title of his PhD project is “Extending lifetime of Norwegian oil installations using predictive maintenance through condition and remaining useful life estimation", and the project is sponsored by OKEA and part of the NTNU program BRU21. Opponents: Harald Rødseth, Ph.D. candidateJon Martin Fordal |
...