Foto del docente

Paolo Castaldi

Associate Professor

Department of Electrical, Electronic, and Information Engineering "Guglielmo Marconi"

Academic discipline: ING-INF/04 Systems and Control Engineering


Keywords: Fault detection and isolation metodologies with application to aerospace system Non Linear Geometric approach Adaptive filtering Longitudinal flight path control in presence of wind shear Guidance Methods Fault Tolerant Control metodologies with application to aerospace systems Fault Detection and Isolation of GPS Fault Detection and Isolation of Aircraft and Spacecraft


The research activities of Paolo Castaldi include:

1.Development of new Fault Detection and Isolation (FDI) methodologies for general aviation aircrafts, UAVs and satellites (2005-today)

2.Development of new Active Fault Tolerant Control (AFTC) methodologies for general aviation aircrafts, UAVs and satellites (2006-today)

3.Development of methodologies for landing in wind shear conditions (2011-today)

4.Development of new methodologies for the identification of static and dynamic Errors-in-variables (EIV) models through the application of ARAIM for GPS/GALILEO (2004-today)

5.Theoretical development and application of new adaptive filtering algorithms for the estimation of the parameters and of the state of induction engines (2004-today)

6.Development and application of Control, Identification and Fault Detection schemes to generic real processes (2005- today (some topic))


•Development of a new linear and polynomial FDI methodology for the aircraft I/O sensors. This approach is based on the development of a residual generator bank and it has been compared with "classical" FDI methods based on residual generator banks developed by means of neural networks or Unknown Input Kalman Filters. Such comparison has been carried out both in case of decoupled and coupled longitudinal and latero-directional dynamics. Particular attention has been paid to methodology robustness: the diagnostic residual generators are decoupled from wind (turbulence and wind gust) and robust against model uncertainties. Finally, the above-described comparison has been made by means of a nonlinear flight simulator including the model of sensors, actuators, turbulence and wind gusts. These simulations have highlighted the superior performances of the proposed approach with respect to the linear techniques presented in literature.

•Development of a new FDI scheme based on the Nonlinear Geometric Approach (NLGA). For the first time in literature a NLGA scheme for a 6 DoF (Degree of Freedom) aircraft model has been proposed, taking also in account the nonlinear engine dynamic and a longitudinal dynamic coupled with the lateral one. This FDI methodology is characterized by residual generators decoupled from wind (turbulence and wind gust) and robust against both critical model errors and parametric uncertainties, due to an inaccurate knowledge of the flight condition. Such decoupling is achieved through the identification of subsystems which are affected by the considered fault and not affected by the other faults and disturbances to be decoupled. On the base of this subsystem it is possible to design residuals which are affected only by the considered fault. The obtained filters have been tested on a flight simulator, including the model of sensors, actuators, turbulence and wind gust model, and compared with filters developed by means of Neural Networks or Unknown Input Kalman Filters. The better achievable performance,s have shown the effectiveness of the proposed algorithm.

•The above described research work has led to the design not only of a FDI unit, but also of a more general FDD (Fault Detection and Diagnosis) unit based on the NLGA (NLGA-FDD). Such unit provides not only the fault detection and isolation, but also the fault estimation in the time domain thus resulting in the development of research topic 2 (see below), i.e. an Active Fault Tolerant Control (AFTC) system based on the NLGA (NLGA-AFTC). In fact, it is possible to achieve the compensation of the fault, affecting satellite or aircraft actuators, by means of a further feedback loop based on the fault estimation. A peculiarity of the proposed scheme is that the fault estimation, thanks to the NLGA, is decoupled from the disturbances affecting the system (wind for aircrafts and gravitational and aerodynamic fields for satellites). Such result allows to obtain an asymptotically unbiased and robust fault estimation, and therefore the rejection of the fault itself.

The NLGA-AFTC approach is conceptually suitable for both aircrafts and spacecrafts and also for generic benchmark systems, as shown in the most recent papers on journals with high IF and invited sessions in international congresses. In particular, by means of the the benchmark, a comparison with Sliding Mode Control (SMC), has shown the better achievable performances of NLGA-AFTC, in particular during the transitory phases following faults occurrence.

•Different NLGA-FDD units decoupled from wind gusts and turbulence have been developed for airplanes and UAVs characterised by 6 DoF models, or UAVs characterised by 3 DoF models. In particular, it has been shown how FDD filters are characterized by different structures depending on the considered model. Nevertheless this investigation has shown how the theoretical base of the NLGA has to be generalized to not affine models. This is a very challenging aspect not present in literature. The achieved results for the NLGA-AFTC have been demonstrated to be better with respect to the ones achievable from other FTC methodologies.

•A NLGA-FTC scheme has been developed also for spacecrafts. In this case, for the first time in literature, the decoupling of the fault estimation (and so of the AFTC scheme) from gravitational and aerodynamic disturbances, particularly significant for Low Orbit satellites (about 500-800 km), has been obtained. A peculiarity of this approach is that, for example air density in case of aerodynamic disturbance, could be unknown. This allows to apply the methodology for planets with uncertain atmospheric information. In particular, it has been shown that the minimum detectable fault size by means of the NLGA-FDD unit is lower than that one obtainble by other methods. This fact allows to design NLGA-AFTC system which allows to reject the fault, even in case where other FTC schemes do not apply fault rejection.

The result of the above described research topics have been presented in many journals with high IF and have several citations. Moreover the topics have been proposed in a large number of regular or invited participations to international congresses. Finally topic 1 and 2 have been the subject of two plenary talk.

The approach has been published in 2 papers and consists in considering the wind-shear (WS) components as faults, and therefore in obtaining WS compensation through an AFTC scheme. The added value with respect to actual literature and the commercial autopilots is important. Fyrther investigation will regard the proof of the stability of the overal control system

Starting from my theoretical results in Errors-In-Variables (EIV) system identification, published on high IF Journal, a diagnosis method has been developed, aiming to detect the presence of faults on pseudorange measurements coming from a GPS/GALILEO system. The EIV approach, assuming that all the variables are affected by errors, allows to obtain more estimations of the same fault. By means of coherency conditions between the different estimations, I have demonstrated how it is possible to identify the fault size and the faulty satellite. The redundancy due to the availability of multiple estimations of the same parameter in the EIV environment lends a greater robustness to the method, compared to the classical FDI methods based on a single least square estimation. This allows to isolate with more confidence the faulty satellite and to obtain a more accurate estimation of the actual position.

With reference to this subject, about which some papers have been published on journals with high IF, it is worth observing that the presented theoretical methodologies have been applied to induction motors. On the other hand , it is wort observing that the theoretical methodology can be applied and generalised to aerospace systems.

•In particular, a new adaptive observer for parameter estimation in an induction motor at standstill has been developed. The method is based on a new canonical form and on a new observer gain law. The main innovation, with reference to the state of the art, lies in the parallel type structure of the observer, particularly robust against noises affecting data.
•Moreover, a new adaptive observer different from the previous one, based on a new canonical form and characterised by a serial-parallel structure, has been developed.

Both the algorithms have been tested with success on the induction motors available at the Laboratorio di Automatica e Robotica (LAR) of the university of Bologna.

Paolo Castaldi has developed and then applied several identification and filtering techniques:

•Blind Identification for communication channels. The usually adopted techniques are based on assumptions not congruent with the real case (assumption of white noises with the same variance affecting the signals). A new methodology has been proposed, which is based on techniques developed for the identification of errors-in-variables models. This extends the results presented in literature since it allows to identify systems which are corrupted by additive noises with different variances and characterised by low signal-to-noise ratios.
•A very important and innovative application, presented on a high IF journal, has been the exploitation of the eigenvector method for the identification of the time constants of the relaxation curve of a NMR signal (Nuclear Magnetic Resonance) in the study of the features of porous mediums.
•An exhaustive investigation on modeling and tracking problems for moving objects in noisy environment has been carried out, in collaboration with ITALTEL s.p.a. (Milan).
•A diagnosis methodology for railroad switch faults has been developed within a CNR project, in collaboration with SABIS s.p.a. (now ALSTOM s.p.s.) in Bologna.