National Network in Applied Mathematics
|
National Network in Applied Mathematics |
|
|
Mathematical Statistics
Centre for Mathematical Sciences
Lund University with Lund Institute of Technology
The programme in Engineering Statistics in Lund is supported by the National Network in Applied Mathematics (NTM), Lund University, Research Councils, NUTEK, STEM, ITM, EU, as well as Swedish and foreign industry. The activity is focused on industrially oriented research in probability and statistics with emphasis on a PhD programme in engineering and industrial mathematical statistics. The support from NTM comes from the subprograms A Stochastic Network (ASN) and the Graduate and Exchange Programme (GEP). It is used mainly as support for PhD students, but PhD advice and workshops are also part of the programme.
Engineering statistics in Lund is focused on three areas of applications:
During the period July 1996 to December 2000 the following people have been involved in the programme, either as PhD students (support from ASN/GEP or other external sponsor is indicated) or as advisors. PhD advisory support from NTM is not indicated individually for the senior researchers.
Senior researchers
Guest researchers
PhD students
The research programme for stochastic methods in telecommunications involves senior researchers professors Søren Asmussen, Tobias Rydén and Jan Holst, PhD students Sofia Andersson (Licentiate 2000), Tiina Heikkinen, Mikael Signahl (Licentiate 2001), and Fredrik Nordström, (Licentiate 2001), PostDoc Francois Roueff (2001). The research concerns both systems and transmission aspects of telecommunication networks, and it is carried out in co-operation with Swedish industry, Ericsson Utvecklings AB, Ericsson Mobile Communication, and Telia Research. Sofia Andersson was employed 1997 and works on traffic modelling and sub-space estimation of Markov models, whereas Mikael Signahl and Fredrik Nordström were employed in 1998 to work on rare event simulation and noise reduction in mobile communication systems, respectively. Tiina Heikkinen was employed as PhD-student in 1999; she works on principles for optimization of communication systems. During 1999 Christian Bergljung, Telia Research, Malmö, spent three months part time with the department under the GEP-programme of NTM.
Traffic in traditional telephone networks was and is successfully described by the Poisson process, but in today's packet switched networks, carrying telephone as well as data traffic and video, traffic patterns are much more complex. Ethernet and Internet are both examples of networks of this kind. Measurements reveal dependencies that the Poisson process is incapable of capturing, whence new versatile traffic models are needed. In addition, there is a need for statistical methods to fit these models to data, and for an understanding of how these methods behave.
Sofia Andersson's PhD project concerns versatile Markovian traffic processes and statistical techniques for fitting them to data. In particular, likelihood-based and sub-space methods are of interest. Markovian models have the feature of enabling analysis of queueing systems using well-established computational techniques, and it is possible to combine simple Markovian models to obtain an overall model which is capable of mimicing the complexity of real data traffic. Likelihood methods are tractable from a statistical point of view, but suffer from computational drawbacks. The optimization process is iterative and the likelihood susrface is typically multimodal. For this reason Sofia also studies sub-space estimation methods, originally designed for linear Gaussian processes, but also applicable in this context. These methods are non-iterative and numerically robust.
Sofia Anderssons work is carried out in co-operation with Ericsson Utvecklings AB (Henrik Nyberg). She took her Licentiate exam in May 2000, when she discussed her thesis: Likelihood-based traffic modelling with Markov-modulated Poisson processes.
Mobile communication is often disturbed by noise and there is need for good denoising algorithms. Several approaches have been tried over the past years, for example spectral subtraction and Kalman filtering. In the work that started during 1997, the wavelet packet transform was used for cleaning noisy speech signals. Different standard wavelet denoising algorithms were used, and a new spectral subtraction approach in the "wavelet packet domain" was presented. We implemented the methods for normal spectral subtraction as well aiming at comparison of methods and development of hybrid algorithms. Bengt Lindoff and Jan Holst have been engaged in this work together with a number of Master Thesis students: Tilak Nissanka, Niclas Svensson, Fredrik Nordström, and Henrik Perbeck. Fredrik Nordström is from the autumn 1998, continuing as PhD student at the department with support from ASN. He works with denoising, and with the symbol interference problem, where interference between different transmitters sending on the same frequency is treated. The work is carried out in co-operation with Ericsson Mobile Communication (Bengt Lindoff).
In a project, temporarily funded by the Institute of Applied Mathematics, Mikael Signahl has started PhD research on rare event simulation. From 1998, Mikael is employed as PhD student in the ASN programme in telecommunications. His project, supervised by Søren Asmussen, aims at the development and study of advanced simulation methods for performance evaluation. Currently a specific problem in assessing bitloss rates in mobile communication is under study in collaboration with Telia Research, Malmö, leading to simulation methodology for Markov chain functionals, mainly in the tail of the stationary distribution. Also some general optimality issues in rare events simulation are under study. Mikael discussed his Licentiate thesis on Derivative estimation via simulation in the presence of discontinuities in January 2001.
Other activities at the department, currently not supported by ASN but within the area of telecommunications, include a PhD project (funished November 2000) on matrix-analytic methods carried out by Jacob Riishede Møller and research on performance analysis for self-similar models carried out by Zbigniew Michna, who recently obtained his PhD degree in this area. Professor Søren Asmussen (queuing theory, stochastic simulation, heavy tails) and professor Tobias Rydén (statistical inference in stochastic processes) are particularly involved in these activities. These two, together with Jan Holst, also act as advisors of the four PhD students on telecommunication.
The research programme in Statistical modelling in energy systems, lead by professor Jan Holst, involves major industries, and is from 1999 also supported by the Swedish Energy Board (STEM) and TFR, as well as by NTM.
Systems for production, distribution and consumption of energy are complex and interwined. Integration of statistical methodology with physical knowledge, measurement techniques, and existing operating experience, is the main research objective in this area. The industrial goals are improved customer comfort and improved reliability in energy delivery, as well as more efficient and environmentally less harmful energy production.
During 1997, two PhD students, Mattias Aronsson and Pia Jonsson, were recruited and partly supported by ASN for the PhD programme for stochastic methods in energy systems. (Both of them also joined the ECMI programme in Lund, started by support from GEP. Pia has, for personal reasons, decided to continue her PhD studies in Kaiserslautern, where she spent the spring semester as an ECMI student. Mattias decided in the fall 1998 to leave for an industrial career.)
In 1998 Roger Halldin was engaged as new PhD student for work on optimization of energy systems, in the project Optimization of power systems under uncertainty. This project is carried out in co-operation with the Division of Optimization and Systems Technology, Royal Institute of Technology, It is supported by ITM and the power industry, i.e. Vattenfall, Sydkraft, Kraftverksföreningen, and Elforsk. Roger Halldin will follow the ECMI curriculum.
This research is supported by STEM, Sydkraft AB, and Malmö Värme AB, through the project Statistical techniques and physical knowledge applied to systems for production, transmission and distribution of district heating. It is a co-operative effort between the department and the Department of Heat and Power Engineering, LTH. Current issues are modelling of power load on district heating networks, climate modelling, modelling of the net from a critical point perspective, model-based predictive control of supply temperature and media flow, optimization of production, net aggregation, alarm prediction aiming at stochastic unit commitment, modelling of hot water load in extreme situations aiming at improved design rules, component supervision, and evaluation and improvement of measuring equipment. These algorithms are included in a new and complimentary system for prediction, control, and supervision of the physical system. The system is developed in co-operation with the operators at the power plant, and the entire process of the development and implementation of a new control system is studied separately as part of the project. Other district heating subjects of interest concern leakage detection and characterization (with Lund Energy) and a study of the relations between optimal performance, operational availability and maintenance.
The work on energy and statistics for improved efficiency and comfort also includes the consumption part. We have during the 1997 and 1998 worked on modelling of heating system components and entire buildings in order to improve on the understanding and performance of the system. Grey Box Modelling, i.e. combination of physical knowledge and statistical techniques, is a major statistical tool used. This work is in co-operation with the Institute for Mathematical Modelling (IMM) at the Danish Technical University (DTU) and Grundfos A/S in Bjerringbro in Denmark. Other partners in the building modelling area are found at the Joint Research Center of the European Commission in Ispra, Italy and in the PASSYS group headed by the Belgian Building Research Institute (BBRI).
The work on electrical power systems resulted during 1997 and 1998 in a number of Master Theses (supervisor Jan Holst). One part of the work concerned robustness in modelling and prediction. The resulting algorithm was implemented in the operation center at Sydkraft and also as part of a power control system offered by an independent consulting company. We have studied nonlinear load modelling including physically based descriptions of the influence from climate and the use of piecewise linear descriptions. One student was awarded Sydkraft's prize for "Best Masters Thesis" at Sydkraft during 1997. In addition we have worked with semi-parametric descriptions of load in order to improve existing linear models with nonlinear, neural network based addenda. The latter work was done by Mattias Aronsson.
During 1997, we initiated a study of the seasonal planning in the Nordic hydrothermal power system, where uncertainties are taken into account and stochastic optimization tools as well as bootstrap methods, are to be used. Roger Halldin, who was recruited as PhD student during the summer of 1998, is working on this project.
This project deals with modelling and prediction of gas loads. The resulting algorithms work with different time steps (hour and day), individually on a number of control stations for monitoring local consumption as well as on a global level. Robustness and high reliability are important features of the algorithm that is under implementation in the control room at Sydgas AB in Malmö. Pia Jonsson (until 1998), Anders Holtsberg and Jan Holst are engaged in this project, which continued during 1999.
In combustion engineering it is of interest to determine properties of flame propagation, such as the flame volume and area at a given time after ignition. To this end, orthogonal image pairs of the flame in a spark ignition engine are obtained. The flame is modelled as a deformed ellipsoid, using Markov Random Fields, and the flame and the cooling surfaces reconstructed using Markov Chain Monte Carlo (MCMC) techniques. The stochastic part of the reconstruction is coupled to the construction of the representation net for the flame surface - this connection has inspired present studies of geometries for MCMC reconstruction. This work is done, in co-operation with the Division of Combustion Engineering at the Department of Heat and Power Engineering, by Finn Lindgren, who was employed as a PhD student during late 1996, and by Jan Holst.
In another co-operative project with Combustion Engineering, we used a wavelet-based method in order to describe the velocity of the gas in the engine cylinder as a function of the piston design. Bengt Lindoff (PhD 1997, now employed by Ericsson) and Magnus Wiktorsson, PhD student at the department, did this work.
The senior researchers in this part of the Engineering statistics programme are professors Igor Rychlik and Georg Lindgren, with contributions also from professors Jan Holst and Ola Hössjer. Seven PhD students, Anastassia Baxevanni, Pär Johannesson (PhD 1999), Ulla Machado, Anders Malmberg, Tord Rikte, Jesper Rydén, and Eva Sjö (PhD 2000), are or have been engaged in the programme. Anastassia Baxevanni came from USA in 2000 with support from Lund university, Ulla Machado, who also follows the ECMI curriculum, was recruited to the programme in 1997 with support from GEP, and Anders Malmberg started 2000 with support from Lund university. External support also comes from TFR, NFR, EU, and ITM, in co-operation with Volvo, Scania, and Swedish Rail. Five foreign PhD students or guest researchers have visited the programme, Per A. Brodtkorb, Department of Marine Hydrodynamics, NTNU, Trondheim, Xavier Pitoiset, Mechanical Engineering and Robotics, Free University of Brussels, Sylvie van Iseghem, IFREMER, Brest, France, Laurence Tual, Vannes, France, and Anastassia Baxevani, IUPUI, Indianapolis.
Pär Johannesson defended his PhD thesis Rainflow Analysis of Switching Markov Loads, on October 29,1999, with professor Harald Krogstad, NTNU, Trondheim as faculty opponent. Pär subsequently has taken up a Postdoc-position at Chalmers, and will spend the year 2000 with Peugeot Citroën in Paris, working on stochastic fatigue. Eva Sjö defended her PhD thesis Crossings and maxima in gaussian fields and seas, on September 8, 2000, with professor Robert Adler, Haiffa, as faculty opponent. After the exam, Eva went to work with Wespot, an Ideon company in Lund.
The scientific questions concern such aspects of stochastic processes which are important for safe and efficient operations of marine and other transportation with the use of modern information technology. Of special interest is the analysis of stochastic aspects of the fatigue process in mechanical structures, in particular analysis of the environmental load and its impact on the fatigue life.
There are three necessary statistical ingredients in any statistical model for fatigue life and safety analysis of large engineering systems: a stochastic load model, including environmental variation and user behavior, a physically relevant and realistic material model which allows material imperfections and variation, and a systems model which describes the response of the system with its implications for systems safety. In Lund, the ASN-supported research is concentrated on the load modelling and evaluation of the load effect.
The research follows two main streams. One is focused on real loads, and the analysis and improvements of methods already used in industry, with the ambition to introduce new methods in practice. Pär Johannesson has been working with load analysis with support from ITM and GEP, and Xavier Pitoiset together with Igor Rychlik on multiaxial fatigue. Xavier defended his PhD thesis in Brussels, 2000, with Igor Rychlik as assistant advisor.
The other aspect is more related to safety analysis and classification of possible damage, for example on ships on duty routes, and systems response, when fatigue is one of the important aspects. On marine applications we co-operate with IFREMER, Brest, and with US Navy. Laurence Tual, Vannes and Anastassia Baxewani, Indianapolis, has worked on this part with the EU-project COMKISS .
The fatigue research is part of the ASN programme on fatigue, which is run jointly between Lund and Stochastic Centre at Chalmers. A Visitors' programme was arranged during March and April 1998 in Göteborg and Lund. Since fatigue is a multidiciplinary field the purpose of the Visitors' programme was to bring distinguished scientists in mechanical engineering from abroad to cooperate with Swedish statisticians and engineers. During March, scientists from Canada, Italy and Japan visited Chalmers and met statisticians from Göteborg, Lund, and Uppsala as well as researchers from different enginering departments at Chalmers and from industry.
An international workshop on Stochastic Methods in Fatigue was arranged at the Swedish Testing and Research Institute (SP) in Borås March 31 - April 2, 1998, with about 80 participants from universities and industry within and outside Sweden. The background of the participants was either in statistics or in engineering and there was a lively discussion about the role and interpretation of statistical models in engineering. The meeting was organized by Boel Wadman and Thomas Svensson from SP, Jacques de Maré and Lennart Josefson from Chalmers, and Georg Lindgren and Igor Rychlik from Lund.
In April the Visitors' programme continued in Lund with participants from Belgium, Canada and Polen as well as from Chalmers, SP and Uppsala.
An intense course in Software for fatigue analysis was held in Lund, January 18-19, 2000, with 15 participants from Swedish industry.
The dynamic stress forces which act on a ship's hull and cause fatigue of its critical members, originate, on one hand, from the regular response of the ship to a random, but rather regularly varying sea, and on the other hand from quite dramatic impacts caused by combination of waves and ship response. The most important of these phenomena is the slamming phenomenon, i.e. the fact that the ships bow rises above the water level and is hit by meeting waves. This has is studied by statistical theoretical methods, in order to predict its frequency of occurrence, and give a statistical characterization of duration and intensity. Researchers involved are Igor Rychlik and PhD students Tord Rikte and Jesper Rydén, in international co-operation with M.R.Leadbetter, University of North Carolina, Krzystof Podgorski, IUPUI, University of Indiana, Indianapolis, and Sylvie van Iseghem, Michel Olagnon, and Marc Prevosto, IFREMER, Brest.
Commonly, the sea is modelled as a Gaussian field with a specified directional spectrum. However, detailed wave observations usually consist of one-dimensional wave records from buoys or travelling ships on route. The problem is to describe and evaluate the statistical properties of waves as objects in space and time, in particular to predict severe and extreme waves. The researchers working on this are Igor Rychlik in co-operation with Krzystof Podgorski, and PhD students Jesper Rydén, Eva Sjö.
Another aspect of extreme waves is that the Gaussian model, derived using linear superposition theory, is less accurate for big waves. Ulla Machado is working as PhD student on statistical wave description for non-Gaussian waves, so called Stokes waves. When a ship moves in a random sea, the deterministic Doppler effect, easily described for simple harmonic waves, becomes a stochastic effect, causing a complicated change in the statistical distribution of encountered waves on the ship. The statistical distribution of the encountered wave period and the Stochastic Doppler shift for a moving point has previously been derived. Ulla Machado now works with the full dynamic response of the whole ship to the random encountered waves. She spent the fall semester 1999 in Trondheim with professor Arvid Naess, Department of Marine Technology, NTNU, and worked with Saddle point methods for computation of Rice's formula, with applications to non-linear waves, wind loads on structures, and slow-drift responses.
Eva Sjö studies the movements of travelling waves and the apparent wave shapes and has in particular studied the intricate problem of joint distribution of wave period and wave length. (The methods will also have applications in Environmental statistics for the study of areas with high concentration of specific pollutants.)
In the EU/Environment project SUCCESS we have been co-operating with marine research institutes and satellite data agencies, for the preparation of a textbook Oceans from Space which appeared in print 1999. The purpose of the project was to make ocean engineers and naval architects more aware of the increasing availability of satellite measurements of the ocean - especially of waves - and how they might incorporate information from these measurements into their designs and it their operational planning.
During 1998 a new EU/Environment project COMKISS was started as a follow up to SUCCESS. The aim of this new project is to demonstrate to three segments of the European maritime industry how satellite derived wave data can be used in design and safe planning of operations. The segments are Shipping Classification, Fast Ship Ferries, and Transportation of Unconventional and Heavy Loads. Within this project Laurence Tual and Anastassia Baxevani has visited the department and worked on statistical description of wave climate along shipping routes. Patrick Brugghe, Breiman Engineering, Holland, visited the department in November 1999 to work together with Igor Rychlik on ship dynamic simulation. The program was finished during 2000.
During a number of years there has been an active co-operation between the department and the Submarine Division at Kockums AB. The efforts have been directed towards target motion analysis (TMA) and navigation.
TMA aims at determining the position of a source of sound from directional and sometimes also frequency data. We have developped a set of algorithms to treat this problem in a changing environment. Some of these algorithms are implemented onboard. We have also, successfully, been working with these kinds of algorithms applied in a fighter plane with infrared directional tracking.
A submarine is equipped with a number of instruments for positioning, such as Inertial Navigation Systems, Gyrocompass, Doppler log etc. We have worked with integration of navigation data from all of these sensors, extending also the model of the dynamic performance of the submarine to high degree of sophistication. This sensor fusion technology is general, with applications e.g. in the process industry and makes improved state estimation, error detection etc possible.
For a complete list of publications for the period 1996-2000, see List of publications
[ Home | Info | People | Research | Education | Consulting | Local Info | Links ]