Research Activity

My main research interests are in the field of model-based diagnosis; in the following I describe in more detail the topics I am focussed on and I provide some related downloads.

Most of these topics are also discussed in my PhD thesis:

PhD Thesis: C.Picardi, Diagnosis: From System Modelling to On-Board Software. 2003.

1. Temporal Aspects

Time is not a secondary issue in diagnosing dynamic systems. I am mainly interested in the use of temporal information in embedded systems (the case study is the electronic control unit of an automotive system). Since diagnostic software is embedded in the device, there are strict requirements both in space and in time; this makes it impossible - at least nowadays - to use a model-based diagnostic engine on-line. My research is focussed on the use of temporal decision trees as a form of on-board diagonstic software. In particular, I have proposed an algorithm which extends the well-known ID3 algorithm in order to generate automatically a temporal decision tree, starting from a fault-symptom table produced by a diagnostic engine.

I am currently working on extending the approach. The first extension I have proposed concerns a fairer evaluation of the tradeoff between losing time and gathering more information during on-line diagnosis. The next step is trying to soften the notion of time in temporal decision trees, with the goal of applying this technique to preventive diagnosis.

Related downloads:

JAIR paper (the algorithm in detail with proofs): L.Console, C.Picardi and D.Theseider Dupré. Temporal Decision Trees: Bringing Model-based Diagnosis on Board. 2003.

LNCS paper (a first extension): C.Picardi. Temporal Decision Trees for Diagnosis:An Extension. 2003.

IJCAI-01 paper (the algorithm, refined): L.Console, C.Picardi and D.Theseider Dupré. Temporal Decision Trees or the Lazy ECU Vindicated. 2001.

DX-00 paper (the initial idea): L.Console, C.Picardi and D.Theseider Dupré. Generating Temporal Decision Trees for Diagnosing Dynamic Systems. 2000.

2. Intelligent Tools for Qualitative Modelling

Most of my research concerns algorithms and tools that work on qualitative models of systems. The need for such qualitative models can actually hinder a wide spread in applicative areas of the methodologies we have adopted. The reasons are that qualitative models are not very common in industry (although they are implicitly used by people carryng out manually diagnostic tasks), and that there are no software tools supporting qualitative modelling.

For these reasons I have devoted part of my research efforts to the analysis and development of intelligent tools for supporting users in building qualitative models. The goal is on the one hand to automate parts of the reasoning process, and on the other hand to help designers in evaluating how a given qualitative model is suitable for the task it has been designed for.

3. System Modelling

Along with my research group I have started an investigation the use of Process Algebras (a formalism widely used in other areas, such as Model Checking or Performance Evaluation) to model systems for diagnosis. Process Algebras allow to describe systems within the same framework many different features of devices, such as single/multiple faults, dynamic/static behavior and time-varying behavior. Currently I am working on the use of Model Checking - applied to Process Algebraic systems- for diagnosis and diagnosability analysis. Another research directions concerns the exploitation of Process Algebras extensions such as timed and probabilistic algebras.

Related downloads:

Artificial Intelligence paper: L. Console, C.Picardi and M.Ribaudo. System Diagnosis using Process Algebras. 2001.

DX-01 (Bridge Workshop) tutorial: C.Picardi and M.Ribaudo. System Diagnosis using Process Algebras. 2001.

ECAI-2000 paper: L.Console, C.Picardi and M.Ribaudo. Diagnosis and Diagnosability Analysis using PEPA. 2000.

DX-00 paper: L.Console, C.Picardi and M.Ribaudo. Diagnosis and Diagnosability Analysis using Process Algebras. 2000.

4. Diagnosability Analysis

Existing systems often do not provide information suitable for diagnosis, because of a misplacement or lack of sensors. This happens because diagnostic aspects are not taken into account during system design, but only afterwards, when the design cannot be modified anymore (at least until the production of a new prototype). Thus diagnosability tests should be integrated in the design process especially for those systems that require on board diagnosis. Here my research focuses mainly on modeling systems in order to test them for diagnosability under different sets of sensors or different models of time, and on integration of such tests in existing design tools.

Related downloads:

DX03 paper (some ideas on generating qualitative models from Matlab(TM)): L. Console, G. Correndo and C. Picardi. Deriving Qualitative Deviations from Matlab Models. 2003.

A paper with CRF about tools for diagnosability analysis: L. Console, G. Correndo, C. Picardi, M. Segnan, R. Bray, A. Buffo, F. Cascio, P. Marchesini. Tools for Integrating Diagnosis in the Design Process. An application to the Common Rail air and fuel delivery systems. 2003.

The IDD paper: C. Picardi, R. Bray, F. Cascio, L. Console, P. Dague, O. Dressler, D. Millet, B. Rehfus, P. Struss, C. Vallée IDD: Integrating Diagnosis in the Design of automotive systems.. 2002.

5. Keystroke Analysis

I have worked with the Security Reasearch Group of our department to develop and experiment a user recognition system based on typing habits. The system was initially based on the analysis of a fixed text typed by the user, but we have extended it so that it can deal with free text, of any length and content. The experiments on free text were concluded a few months ago so... check here for upcoming results!

Related downloads:

ACM TISSEC paper (the first recognition system, with fixed text): F. Bergadano, D. Gunetti and C. Picardi. User Authentication through Keystroke Dynamics. 2002.

IDA paper (results on dynamic keystroke analysis that lay the grounds for free text analysis): F. Bergadano, D. Gunetti and C. Picardi. Identity Verification through Dynamic Keystroke Analysis. 2003.