RICERCA ITALIANA     

Research program

Integrated Methods and Algorithms for NonDestructive Evaluations of architectural heritage

University Co-ordinator

Politecnico di BARI - ELETTROTECNICA ED ELETTRONICA - BARI(BA)

Research Unit Leader

Giuseppe ACCIANI

Description

The evaluation of both the preservation state and the vulnerability of artistic and architectural heritage with respect to structural risks, particularly the seismic one, is a very important issue in our country. Well-established methods to evaluate mechanical and structural characteristics of ferro-concrete and steel works cannot be used to analyze masonries, since in our country there are many different masonries from a typological point of view as well as from a constructive one. Thus, dedicated (non-destructive or very little invasive) techniques must be used to evaluate the risks of both static instabilities and collapses due to a seism.
At the moment several non destructive diagnostic techniques are available, and some of them are even applied to historical buildings. Nevertheless the evaluation of the vulnerability state is often made on the basis of incomplete or uncertain information; in such cases expertise has an important role in evaluating the available indexes and obtaining final information.
The proposed research starts from the acquisition, fusion and processing of signals coming from several sensors. The goal will be the development of an integrated diagnostic system for masonries, based on several non-invasive investigations and their interpretation based on cooperative clustering. The final step would be the identification of the most suitable recovery methodologies.
The research group 1 will focus on feature extraction techniques and multi-sensor signal classification. More precisely, both static and dynamic thermo-graphic images of masonries will be acquired and analyzed. The group has a thermo-camera IR "TVS700", already employed for masonry analysis in cooperation with the ICAR 09 group. Thus, a considerable amount of images are available.
Furthermore, classification and analysis will be performed on sonic, ultra-sonic and electromagnetic measurements supplied by the other research groups.
In the last years the research group 1 has acquired specific knowledge on the following issues: clustering, supervised and unsupervised, of vector data; processing of mono- and multi-dimensional signals, of video sequences and fixed images; development of neural units devoted to the realization of both fuzzy and crisp neural networks for the classification of multi-dimensional data; feature extraction techniques from multi-dimensional data structures for signal classification.
The research results have been recently applied to industrial non-invasive diagnosis of both finished and unfinished industrial products.
Moreover, the experience on multi-dimensional signal processing and on band decomposition techniques of mono- and multi-dimensional signals (wavelet, sub-band coding) has been used in both pattern recognition and feature extraction techniques for non-invasive diagnosis of thermo-graphic video sequences of composite materials.
Goal
One of the advantages of an integrated multi-sensor technique is to join the characteristics of different kind of measurements coming from different cooperating systems in a single experimental tool. A single system could process the measurements gathered by several sensors better than systems which use the information of a single measurement. In fact, the integrated system can take into account in different way the data of each sensor, because even if the information coming from a single sensor could be sufficient to describe the problem under test, the information present in the other measurements can complete the knowledge of the physical/geometrical characteristics in the same problem.
In this analysis the data pre-processing is a very important phase. In this stage, the measurements of each sensor must be arranged into vectors that can be analyzed by the integrated system. The aim of the analysis is to identify the type of the defect (crack, fissure, bubble, cavity, rust etc.), its location and size.
In the recent years, IR thermography has been widely employed to inspect bridges and ferroconcrete buildings. Furthermore it can be used for the diagnosis of historical buildings because of two advantages: it is non-invasive and easy to apply. Moreover it allows evaluating the static behavior of masonries with hidden restores. In fact, it is possible to have images of the layers under the plaster without removing it, which is very important when a fresco is present.
The thermography-based diagnosis consists of a false colors image which displays the temperature map of the object under test. The images can be acquired in a static way only when one image shows the reflectivity or steady state thermal conductivity; the acquisition is dynamic when a sequence of images displays the reflectivity of a masonry which cools itself after heating.
The first acquisition technique is very useful to identify the so-called thermal bridges. Thermal bridges are spatial regions with high superficial heat draining due to high thermal conductivity. The second acquisition is employed to determine the defects of the composite materials. Non homogeneous parts of the superficial layers can be highlighted since the lack of homogeneity changes locally the propagation of the thermal phenomenon. The thermographic techniques create data with high spatial resolution, but generally they are effective only for thin walls (superficial images with dynamic thermographic images).
Project steps
Step1 Thermographic Analysis
In this stage the unit 1 will deal with thermografic investigations. The acquisition methodology and the peculiarity of the measurement tools will be studied. A suitable procedures to obtain correct measurements of thermal signals depending on the kind of defect will be identified.
Each type of analysis highlights some specific features of the structure under test. Thermographic images allow to evaluate mainly two parameters: the thermal reflectivity of the superficial layer (dynamic acquisition); the mean thermal conductivity of the structure under test (static acquisition).
Finally the types of masonry and their defects will be identified in order to provide the most suitable methodologies to detect the defects. This aspect will be carried on together with the colleagues of Architecture Department of Politecnico di Bari. Therefore the most detectable defects will be individuated by means of thermographic techniques.
This phase will last about three months.
Step 2: Acquisition, Data pre-processing and Feature Extraction
In this phase the most suitable acquisition and pre-processing methods will be found in order to obtain the most meaningful information to study the masonry. Aim of this phase is to process the information coming from sensor measurements and to achieve low dimensional data that can be easily used in the cooperative clustering stage.
The pre-processing will have as a main purpose the choice of the algorithm able to reduce the information characterized by features or classes of defects. Each thermographic technique (static or dynamic) needs different systems to point out the relievable classes of defects in the best way.
In this step neuro-fuzzy network based systems or standard analysis techniques could be used. Moreover decomposition techniques in sub-band or wavelet could be adopted because they can underline the texture in the images of the defects.
This step will last eight/ten months.
Step 3: Multisensor Data Collection
The cooperative processing of signals needs an alignment stage if the signals come from sensors of different nature. The first stage of this step will be the study of the alignment technique of the data provided by the research project units. In this stage a structure of test will be identified. Exploiting this structure each research unit will be able to take measurements related to a specific technique and than pre-process the data.
The pre-processed information will be projected onto the multi-dimensional grid which will be able to describe the structure under test, and will be stored in a central data-base with direct access of each research unit. Starting from the multi-dimensional data, a first analysis stage will compare the information inside each type of data to the different defects. This step will be led by exploiting the experience of the colleagues of ICAR 09 group.
This phase will take three or four months.
Step 4: Clustering of the multi-dimensional data.
In this phase the research group 1 will be involved in processing of the data deriving from the sensors applied to the structures under test. The aim of this step will be the design and development of processing techniques for multi-sensorial data in order to extract the classes able to characterize the preservation state of the masonry. The availability of a large quantity of heterogeneous data referred to the same system allows for many research tasks to be investigated. In this phase, in order to better individuate proper processing techniques for information extraction, the previous techniques already utilized for thermographic images will be used for multidimensional data analysis too.
The procedures will use:
Clustering, supervised and unsupervised, or techniques based on the information decomposition in statistically uncorrelated terms (PCA), or independent terms (ICA). These creates maximum correlation data domains, which are successively classified with crisp techniques (based on distance measures of a point from the prototype of a fixed class) or with fuzzy ones.
A-priori information, related to the identifying of specific features in data and based on the experience of experts in architectural area. In this type of investigation, the most used techniques use supervised systems, in which the expertise is ‘frozen' in the data processing parameters.

In this step, after a complete analysis, it might be necessary to reconsider one or more pre-processing techniques applied to the data deriving from one or more sensors developed by each research group, in order to enhance the information quality. This step will be conducted in the second year until the end of the research project and together with the definition of the collaborative clustering techniques described in step 5.
Step 5: Definition of an intelligent technique for classification of multisensorial and multidimensional data, based on technique of cooperative clustering.
In this last phase, all the results obtained from the previous study will be gathered and a clustering cooperative structure will be defined. In fact, all the pre-processing techniques adopted for each sensor and the several typologies of both supervised and unsupervised clustering can coexist. Particularly, the research unit 1 will design a cooperative clustering technique, specific for data deriving from different types of sensors and characterized by different granularity.
These techniques have a double advantage:
to incorporate expertise into the automatic system;
to avoid to excessively constrain the definition of the classes which represent the structure of the available data.

In this phase the comparison with the results obtained by the research unit 4 will be essential in order to define the best methodology for data processing.
This part will be developed in the last six months of the research project.