M.Sc. thesis proposals

Control of timed discrete event systems with timed specifications

• Discrete event systems (DES) are characterized by a discrete set of states, and state changes are driven by the occurrence of (asynchronous) discrete events. This is a very general class of models, which is often encountered in automated manufacturing, communication and transportation networks, control applications, etc.
• In timed DES the transitions between activities are further endowed with time bounds in order to represent the duration of activities. A timed control task involves reaching specific states (while avoiding others) within prescribed time limits.
• The design of a controller for timed DES is a complex and challenging task that requires the development of suitable tools for reachability analysis, path optimization, and control policy design. This thesis aims at the development of a Matlab library of such tools for time Petri nets and its testing on some timed DES control benchmarks.

Prediction of stuck-pipe events in oil&gas drilling operations

• Drilling is a complex process that consists in boring a hole (that can be several thousands of meters long) in order to create a well for oil (or gas) production. A sticking occurs when the drillstring cannot be neither rotated nor moved along the axis of the wellbore. A pipe is considered stuck if it cannot be freed from the hole without damaging the pipe, and without exceeding the drilling rig's maximum allowed hook load. This phenomenon can be due to pressure-related issues (if the pressure in the annulus exceeds that in the formation being drilled, the drillstring is pulled against the wall of the borehole and held against it) or to mechanical blocking (e.g., due to incomplete removal of the cuttings, or partial collapse of the borehole). Stuck-pipe phenomena can have disastrous effects on drilling performance, with outcomes that may range from time delays (note that a day of drilling can cost millions of euros) to loss of expensive machinery. This thesis aims at the development (using machine learning techniques) of a data-based prediction model that can be used to anticipate the occurrence of the mentioned events. A large dataset of oil&gas wells can be used to train and validate the model.
• This work is part of an ongoing collaboration with Eni's Drilling & Completion Risk Analysis Department.
• The research team includes Prof. Matteo Matteucci and Dr. Federico Bianchi.

Data-driven prediction of paper break events in paper manufacturing process

• During the paper manufacturing process, some paper break events can occur. If a break happens, the entire process is stopped, any found problem is fixed and the production is resumed. Resuming the process can take more than an hour, which represents a severe cost for the factory. Even a 5% reduction in the break events will yield a significant cost saving. The objective of this thesis is the development of suitable techniques to predict such breaks in advance (early prediction) and identify the potential cause(s) to prevent the break. To build such a prediction model, the applicant will use data collected from a network of sensors deployed to monitor a real manufacturing plant.
• The research team includes Dr. Federico Bianchi.

Identification of hybrid systems

• Hybrid systems are heterogeneous dynamical systems whose behavior can be described by the interaction of several time-driven continuous dynamics (continuous states or modes) indexed by event-driven discrete dynamic (discrete states). They can represent technological systems where continuous models such as differential or difference equations describe the physical and mechanical part, and discrete models such as finite-state machines or Petri nets describe the software and logical behavior. For example, while driving, a car follows different dynamics, depending on the specific phase (acceleration/braking, cornering) or the track characteristics (slope, road surface, weather conditions).
• The goal of this thesis is the development of a black-box technique to identify both the system modes and the switching mechanism directly from data.
• The research team includes Dr. Federico Bianchi.

Adaptive Kalman Filtering for Hybrid Systems
(in collaboration with Prof. Simone Formentin and Dr. Federico Bianchi)

• The filtering problem consists in reconstructing or estimating a non-measurable variable (called remote variable) based on the knowledge of other, measurable and uncertain variables. This uncertainty refers to measurement noise related to sensors, and to process noise which accounts for all those modelling aspects that for various reasons are not taken into account. In Kalman filtering, the noise is expressed by means of stochastic Gaussian processes described in terms of their statistical properties (mean and covariance). The filter formulation requires the knowledge of a model of the underlying system, as well the process and measurement noise covariance matrices. In practice, the latter are design parameters, whose setting critically affects the filter performance.
• This is all the more true for hybrid systems, whose behavior can be described by the interaction of several time-driven continuous dynamics (continuous states or modes) indexed by event-driven discrete dynamic (discrete states). Indeed, in that framework also the noise covariances change with the discrete state.
• The problem of estimating the covariance matrices that optimize the filter performance is an open problem in literature, which could have a significant impact on Kalman filter-based applications.
• The goal of this thesis is the investigation of the filtering problem in the context of hybrid systems, with particular emphasis on the adaptation of the covariance matrices involved in the filter computations.

Methods for motorcycle attitude estimation
(in collaboration with Prof. Simone Formentin and Dr. Federico Bianchi)

• Nowadays, wheeled vehicles are equipped with many driver assistant systems - such as the ABS (anti-lock break system) or the TCS (traction control system) - which increase driver and passengers safety and comfort. While these systems have become a standard on all four-wheeled vehicles, their use on two-wheeled vehicles is more recent (the ABS became mandatory in the EU only in 2016). Therefore, in the field of two-wheeled vehicles, the interest in electronic systems for the control of vehicle dynamics is still recent.
• Particularly interesting and challenging is the attitude estimation, since it plays a major role in the overall vehicle dynamics. In this regard, a reliable method should take into account also the multi-mode nature of the controlled system: the overall dynamics can be considered as a collection of several continuous dynamics representing different operational modes - acceleration, braking, steering - indexed by a discrete variable.
• The goal of this thesis is the investigation and development of methods for motorcycle attitude estimation.