Computational Intelligence Methods & Decision Support Tools in Cultural Materials Anastasios Doulamis, Anastasia Kioussi, Maria Karoglou, Klio Lakiotaki, Ekaterini Delegou, Nikolaos Matsatsinis and Antonia Moropoulou National Technical University of Athens National Technical University of Athens, School of Chemical Engineering Computational Intelligence & Decision Support Tools • Assist experts to take solid decisions • Reject non-preferable solutions – Reduces the costs • Identify hidden knowledge – Image processing/analysis, computer vision • Improve validation performances • Results on optimal consolidation of cultural material National Technical University of Athens, School of Chemical Engineering Outlines National Technical University of Athens, School of Chemical Engineering Optimal Consolidation of Cultural Heritage Material • Cultural heritage protection => targeted restoration actions to increase monuments’ lifetime. – conservation materials • The performance of each material on the restoration significantly differs with respect to its type, chemical properties and the building substrate. • Design phase: A decision support system which will assist the engineer to extract optimal conclusions. • Today, such section is expert-dependent process mainly exploiting her/his experience. National Technical University of Athens, School of Chemical Engineering Computational Intelligence in Cultural Material Consolidation • We have applied different types of intelligent tools for optimal selecting the most suitable conservation materials Linear regression Supervised Neural Intelligence National Technical University of Athens, School of Chemical Engineering Fuzzy Kmeans & Neural Intelligence AI & DSS for Conservation Interventions • Applications to two Cases of Conservation Interventions – Consolidation of Materials/Structures – Cleaning of architectural surfaces National Technical University of Athens, School of Chemical Engineering Consolidation interventions How to support the decision making in choosing the most appropriate consolidation material The consolidation materials and interventions used intend to the : Modification of micro structural characteristics of the stone, leading to lessening of stone susceptibility to salt decay Prevention of decay due to grains de-cohesion Amelioration of mechanical characteristics of the stone Main categories of consolidation products are: • Inorganic Materials • Nano-limes • Organic Materials • Alkoxysilanes National Technical University of Athens, School of Chemical Engineering Validation in Lab and in the Monument Scale National Technical University of Athens, School of Chemical Engineering DDS OUTPUT Parameter Climax Availability in Greece Parameter Penetration depth Unknown / Not satisfactory / depend on any factors/ depends on the solvent / good / very good with the use of specific solvent / very good/ Color change No change /depend on the surface / depend on the excess of material at surface/ medium / high/unknown Yes/No / Unknown Irreversibility 0-10 Durability at environmental loads Chemical compatibility Climax 0-10 Capillary absorption of water Low / medium / high/unknown 0-10 Change of hardness Low / medium / high/unknown Standards Yes/No / Unknown Creation of film 0/1/2 National Technical University of Athens, School of Chemical Engineering Criteria Adopted Criteria Adopted Availability Inversibility Resistance to Chemical Environmental Compatibility Standardized Rules Loads Yes/No Numerical Numerical Numerical /Unknown Values Values Values Filming Penetration Discolored Water depth Numerical Quality Values Values National Technical University of Athens, School of Chemical Engineering Binary Values Hardness absorption Quality Values Quality Values Quality Values Experiments Set-Up • Two scenarios for different application substrate: • The ranking is primarily based on chemical composition of stones 1st Scenario: Calcareous Stone (35 samples) 2nd Scenario: Silicon-based Stone (34 samples) In future, more parameters will be included like micro-structural characteristics of material, mechanical properties etc, National Technical University of Athens, School of Chemical Engineering A Feed-Forward Neural Network w10,1 w11 w10,2 w10,q w20,1 Input Vector .. . w21 w20,2 w20,q w 0J ,1 .. . z wq1 w 0J ,2 w 0J ,q wi0 wi0,1 , wi0,2 wi0,q w1 w11 , w21 wq1 National Technical University of Athens, School of Chemical Engineering T T i 1,2 J Neural Networks Set-Up • Three categories (preferred, neutral non-preferred) • Two categories (preferred, nonpreferred) Continuous Outputs • Preference Order Quantized Outputs National Technical University of Athens, School of Chemical Engineering • Two layers networks • Constructive training method Different Network Sizes Results- Generic 2 1.4 Neural Output in Training Samples Neural Output inTesting Samples Desired output in Training Samples Desired Output inTesting Samples 1.5 1.2 1 0.5 Selection Preference Selection Preference 1 0 0.8 0.6 -0.5 0.4 -1 0.2 -1.5 0 5 10 15 20 25 Number of Samples 0 1 2 3 Average Error over 10 randomly partitioned datasets when 20 hidden neurons are selected Training error Testing Error 1.5% 23.7% National Technical University of Athens, School of Chemical Engineering 4 5 Number of Samples 6 7 8 Effect of Network Size-Generic 55 50 Error in Testing Dataset (%) 45 40 35 30 25 20 2 4 6 8 10 12 Network Size National Technical University of Athens, School of Chemical Engineering 14 16 18 20 Results- Scenario 1 1.5 Neural Output inTesting Samples Desired Output inTesting Samples Selection Preference 1 0.5 0 -0.5 -1 1 test_output 2 3 4 5 Number of Samples 6 7 Average Error over 70 randomly partitioned datasets when 20 hidden neurons are selected National Technical University of Athens, School of Chemical Engineering Training error Testing Error 1.2% 22.4% Combined Fuzzy K-means & Neural Networks Testing on Data Neural Networks (supervise d) Fuzzy K_Means (unsupervised) National Technical University of Athens, School of Chemical Engineering Results –Scenarios 1,2 Average Error over 70 randomly partitioned datasets when 20 hidden neurons are selected Average Error over 70 randomly partitioned datasets when 20 hidden neurons are selected SCENARIO 1 SCENARIO 2 Training error Testing Error 0.4% 5.7% National Technical University of Athens, School of Chemical Engineering Training error Testing Error 0.7% 2.6% The UTA* algorithm AR Alternative Reference set Acronal 500 D Conservare® H 100 Ludox HS30 Mowilith 30 Paraloid B72 g1 g2 g3 Reversibility Availability Hardness 6 yes 2 6 3 4 5 Ranking order 1 wij ui ( gij 1 ) ui ( gij ) 0, g4 low Penetration depth good k=1 no low satisfactory k=2 7 2 no yes medium large good dependent k=3 k=4 2 i=1 yes i=2 7large i=3 Very good i=4 k=5 i 1, 2,..., n [ ( k ) ( k )] [min] z 1 subject to (a , a ) if ak ak 1 k k 1 k ( a , a ) 0 if a a k k 1 k k 1 n ai 1 wij 1 i 1 j 1 wij 0, ( k ) 0, ( k ) 0 i, j and k and j 1, 2,..., ai 1 ui ( g1i ) 0 i 1, 2,..., n j 1 j wit i 1, 2,..., n and j 2,3,..., ai 1 ui ( gi ) t 1 (ak , ak 1) u[g(ak )] ( ) ( ) u[g(a 1)] ( 1) ( 1) ai 1 * National Technical University ofu Athens, Post-optimality analysis i ( gi ) wij i 1,2,..., n School of Chemical Engineering j 1 m [ (a ) (a )] z k k 1 Criteria weights * k 19 Results-Scenario 1 Criteria Adopted Availability Inversibility Resistance to Chemical Standardize Compatibili d Rules Environmen ty tal Loads Yes/No Numerical Numerical Numerical Binary /Unknown Values Values Values Values Filming Penetration Discolored Water Hardness depth absorption Numerical Quality Quality Quality Quality Values Values Values Values Values National Technical University of Athens, School of Chemical Engineering Results-Scenario 2 Scenario 1 0.0085 0.120704544 0.054324815 0.096816889 0.097848487 availability in Greece reversibility 0.120795954 0.067767522 Scenario 2 0.062514711 0.10829301 durability 0.09743482 chemical compatibility durability 0.114843951 0.103792321 chemical compatibility standards film film penetration depth 0.030142594 color change 0.299685734 reversibility standards penetration depth 0.088442145 0.048352749 availability in Greece water absorption 0.090789818 National Technical University of Athens, School of Chemical Engineering color change 0.059719141 0.323630796 water absorption hardness Porous Biocalcarenite Pilot scale treatments for porous stone consolidation in the Medieval City of Rhodes Materials LUDOX HS30 (PL) Silbond HT20 (PH) Rhodorsil RC70 (RP) Acryl Siliconic Resin (EU) National Technical University of Athens, School of Chemical Engineering Evaluation of the Compatibility of Conservation Interventions in lab Water Absorption of Porous Stone & Consolidated Porous Stones 25 Imbibed Water, wgt. % RPS CSPH2 20 CSPL3 15 10 CSEU1 5 CSRP4 0 0 10 20 30 40 50 60 70 80 (Time, sec) 1/2 Changes of water absorption curves (capillary) of consolidated porous National Technical University of Athens, stones monitoring by infrared thermography in the laboratory School of and Chemical Engineering IR Thermography Investigation of Capillary Rise, Monument Scale Investigated Surface: Gate of St. Paul, Medieval Fortifications of Rhodes Evaluation of Pilot Consolidation Interventions, Monument Scale Investigated Surface: Entrance of Moat, Medieval Fortifications of Rhodes 15 months after the applications 28 months after the applications Consolidation Materials: LUDOX HS30 (PL), Silbond HT20 (PH), Rhodorsil RC70 (RP) acryl siliconic resin (EU) University of Athens, National Technical School of Chemical Engineering Validation of the results -in laboratory (various analytical techniques like capillary absorption test, mercury intrusion porosimetry etc) -in monument scale (non-destructive testing) Feedback: Changes in materials ranking National Technical University of Athens, School of Chemical Engineering Computational Intelligence on Cleaning Interventions Assessment • We have applied the aforementioned methodology for supporting the decision making on the assessment of pilot cleaning interventions on marbles surfaces • The application sites are located on the historic buildings of National Library of Greece (NLG), and National Archaeological Museum in Athens-Greece (NAM) • The diagnosed decay patterns are black grey crusts, washed out surfaces, and fractured surfaces of marble. National Technical University of Athens, School of Chemical Engineering Presentation of Applications Sites Smooth Marble Architectural Surface NLG facade Different protection degree from rain wash North Facade National Technical University of Athens, School of Chemical Engineering Presentation of Applications Sites Relief Marble Architectural Surface East Side, Full Protected from rain Wash NAM West Façade, North Column Relief Marble Architectural Surface Column Capital East Side, Full Protected from rain Wash Relief Marble Architectural Surface North Side, Different Protection Degree from Rain Wash National Technical University of Athens, School of Chemical Engineering North Column East Side Decay Diagnosis - NLG Depending on the protection degree from the projected horizontal geison FOM Friable black-grey crust Cohesive black-grey crust Inter-granular fissured marble National Technical University of Athens, School of Chemical Engineering SEM FTIR Decay Diagnosis - NAM FOM SEM Black-grey crust East orientation FTIR Washed out surfaces FOM SEM North orientation FTIR National Technical University of Athens, School of Chemical Engineering Decay Diagnosis - NAM Black-grey crust of great variety FOM regarding: the width, the presence or not of FOM SEM Barite (patina), the location in the anthemia relief, relief side East orientation Relief face right part FTIR Relief side, central part Relief face central part FOM FOM SEM FTIR National Technical University of Athens, School of Chemical Engineering FTIR SEM In situ application of pilot cleaning interventions – Monument scale - NLG Smooth marble architectural surfaces, black-grey crust, inter-granular fissured marble Pab22, P. ΑΒ57, 2h Pnc22, P. (NH4)2CO3, 2h Pm22, P. Mora, 2h Ps32 P. Sepiolite, 3.5h Pat2, Atomized water Pm24, P. Mora, 1.5h Ps34 P. Sepiolite, 3h Pab24, Pnc24, National P. Technical University of Athens, ΑΒ57, 1,5h P. (NH4)2CO3, 1.5h School of Chemical Engineering In situ application of pilot cleaning interventions – Monument scale - NLG Smooth marble architectural surfaces, black-grey crust, inter-granular fissured marble Pnc12, P. (NH4)2CO3, 1h Pab12, P. ΑΒ57, 1h Pab14, P. ΑΒ57, 1h National Technical University of Athens, Pnc14, P. (NH4)2CO3, 1h School of Chemical Engineering Ped12, Π. EDTA, 1h Ped14, P. EDTA, 1h In situ application of pilot cleaning interventions – Monument scale - NAM Relief marble architectural surfaces, black-grey crust, east orientation Ke Ke Ion exchange resin with solution of (NH4)2CO3, 40min Ke2a Ke3b Π. ΑΒ57, 5min Ion exchange resin with deionised water, 30min Biological poultice Ke Ke1c Wet micro-blasting method •Spherical particles of CaCO3 d<80μm, •Function pressure 0.5bar, •Proportion of CaCO3/water: 1/3, •d nozzle 12mm • working distance 50 cm University of Athens, National Technical School of Chemical Engineering KeG3 In situ application of pilot cleaning interventions – Monument scale - NAM Relief marble architectural surfaces, washed out surfaces, north orientation Kn P. ΑΒ57, 5min Kn3c Ion exchange resin with deionized water, 40min Kn3b Ion exchange resin with deionized water, 10min Kn2a Ion exchange resin with solution of (NH4)2CO3, 20min Kn1a Ion exchange resin with solution of (NH4)2CO3, 10min Kn1b Double application of Ion exchange resin with solution of(NH4)2CO3, 2x10min Kn1c Kn2b Ion exchange resin with deionized water, 20min Kn2c Double application of Ion exchange resin with deionized water, 2x20min Kn2d National Technical University of Athens, School of Chemical Engineering Wet microblasting method In situ application of pilot cleaning interventions – Monument scale - NAM Relief marble architectural surfaces of capital, east orientation Kke Ion exchange resin with solution of (NH4)2CO3, 10min Ion exchange resin with solution of (NH4)2CO3, 40min Ion exchange resin with deionized water, 60min Kke1a Kke2a Kke3a1 Kke3a2 Ion exchange resin with deionized water, 10min Ion exchange resin with deionized water, 20min Wet micro-blasting method kkeg51 Kke1b Ion exchange resin with of Athens, National Technical University deionized water, School of Chemical Engineering 30min Kke2b Kke3b Ion exchange resin with solution of (NH4)2CO3, 20min kkee3ba P. ΑΒ57, 5+ 15 min Assessment of Cleaning Interventions: Techniques & Parameters Applying, after cleaning the same experimental techniques that were applied before cleaning, a methodological approach for cleaning assessment is compiled. Comparison of the marble surfaces physico-chemical characteristics before & after cleaning, along with recording the variations of the corresponding critical parameters, makes feasible the recommendation of the best cleaning according to the examined case. Digital Image Processing of SEM images: fracturing of the surface Shape factor (a roughness factor) SEM-EDS: chemical & mineralogical composition stratification, total crust width, patina, macro-crystalline gypsum layer width, micro-crystalline gypsum layer width Fracture Density Patina preservation index Friability index Preservation index of gypsum layer National Technical University of Athens, School of Chemical Engineering Assessment of Cleaning Interventions: Techniques & Parameters Laser Profilometry: texture & roughness assessment Roughness Rq (μm) Surface area, (ratio of actual to projected area) Colorimetry CIELab color space: evaluation of color modifications L, Luminosity total colour difference ΔE difference in difference in red-green blue-yellow a* Technical University of Athens, National b* School of Chemical Engineering Assessment Criteria & Critical Assessment Parameters of Cleaning Interventions – Experimental Techniques Cleaning Assessment Criteria Chemical-mineralogical composition of the surfacesstratification Preservation of Patina, Preservation of Authentic Material, Scanning Electron Microscopy with Energy Dispersive Xray Spectroscopy Color Texture, Morphology & Surface Cohesion - Surface Microstructure Removal of Black Depositions Surface Preservation State – Decay Susceptibility – Durability Roughness, Rq, Fracture Density Ratio of actual to projected area - Surface area Colorimetry Laser Profilometry Digital Image Analysis of Scanning Electron Microscopy Images Experimental Techniques for Measuring Critical Assessment Parameters of Cleaning National Technical University of Athens, Critical Assessment Parameters of Cleaning School of Chemical Engineering Total Color Difference, ΔΕ Colorimetry Monitoring of Surface Preservation State – Durability of Marble 1. 2. Decay patterns distribution on the building are mainly controlled by material location, orientation, protection from rain-wash, atmospheric conditions and pollution. However, the long-term aesthetical and structural properties of marble are closely related to the lateral and vertical distribution of particulate matter and salts-gypsum, as well as to the bonding of the calcite grains in the matrix; factors that are strongly affected by cleaning. Therefore there is an urgent need for a tool to interrelate information – data, between space and physical-chemical characteristics of building materialsmarble, taking into account their variation over time. The suggested methodological tool is a GIS platform National Technical University of Athens, School of Chemical Engineering Materials Mapping in GIS, Façade, National Archaeological Museum Working scale: building’s facade materials mapping performed using GIS based on the acquired data by NDT and inlab analytical techniques. Acropolis Athens The area extend of each investigated material was calculated by the means ofof GIS Historic plaster area was 248.56m2, whereas new plaster area was 13.63m2 National Technical University of Athens, School of Chemical Engineering Materials Mapping in GIS, Façade detail, National Archaeological Museum Working scale: building’s facade Digital decay mapping performed using GIS based on the acquired data by NDT and in-lab analytical techniques. Brown color depicts areas of coating total detachment and intense fracturing total area on west façade: 25.56m2 Acropolis Blue color represents the areas of coating loose interface to the substrate total area on westof Athens façade: 219.72m2 National Technical University of Athens, School of Chemical Engineering Façade detail - National Archaeological Museum •material type •applied cleaning method •application details •application area •cost National Technical University of Athens, School of Chemical Engineering Acropolis of Athens Working scale: building’s facade GIS thematic maps for decay & pilot conservation interventions Recording & ascribing attributes to features Attribute db, (physical-chemical data, indexes of building material preservation state, before and after conservation) GIS db, (topological characteristics like area, perimeter, adjacency, etc) Relational Data Base National Technical University of Athens, School of Chemical Engineering Spatial Classification of Decay. Different Physical-chemical characteristics & spatial properties Decay thematic map for the capital surface, along with RDBs for both front & side anthemia surfaces RDB RDB Spatial Classification of Conservation Interventions Pilot conservation interventions’ thematic map for the capital surface, along with the RDB of the front anthemia surface RDB GIS analysis using Boolean and logical operations on decay thematic map for the capital surface Spatial entity in compliance with the combined expression central area of the anthemia relief Which is the entity that comply with: 1) roughness ≥ 7, 2) fracture density ≥ 35.3 RDB Suggested Information Management System ANALYSIS & OPERATIONAL TOOLS RELATIONAL DATABASE (ATTRIBUTES) GIS SPATIAL DATA Using the continuous process of GIS platform datasets concerning building pathology & conservation interventions are recorded, correlated, distributed & attributed to space in different working scales during different time periods Support on decision making for cleaning assessment using Computational Intelligence Results-Crust 4.5 3.5 KMeans Output inTesting Samples Desired Output inTesting Samples 4 3 3 Selection Preference Selection Preference 3.5 2.5 2 2.5 2 1.5 1.5 1 0.5 0 2 4 6 8 10 Number of Samples 12 14 16 18 National Technical University of Athens, School of Chemical Engineering 1 1 Neural Output inTesting Samples Desired Output inTesting Samples 1.5 2 2.5 3 3.5 4 Number of Samples 4.5 5 5.5 6 Results-Washed-out 4.5 4.5 KMeans Output inTesting Samples Desired Output inTesting Samples 3.5 3.5 3 3 2.5 2 2.5 2 1.5 1.5 1 1 0.5 1 Neural Output inTesting Samples Desired Output inTesting Samples 4 Selection Preference Selection Preference 4 2 3 4 5 Number of Samples 6 7 8 9 National Technical University of Athens, School of Chemical Engineering 0.5 1 1.2 1.4 1.6 1.8 2 2.2 Number of Samples 2.4 2.6 2.8 3 Results-Fractured 4.5 4.5 KMeans Output inTesting Samples Desired Output inTesting Samples 4 4 3.5 3.5 3 Selection Preference Selection Preference KMeans Output inTesting Samples Desired Output inTesting Samples 2.5 2 3 2.5 2 1.5 1.5 1 1 0.5 1 2 3 4 Number of Samples 5 6 7 National Technical University of Athens, School of Chemical Engineering 0.5 1 1.2 1.4 1.6 1.8 2 2.2 Number of Samples 2.4 2.6 2.8 3 Results-Overall Average Error over 100 randomly Average Error over 100 randomly partitioned datasets when 20 hidden neurons are selected partitioned datasets when 20 hidden neurons are selected CRUST FRACTURED Training error Testing Error Training error Testing Error 5.3% 18.5% 3.0% 11.2% Average Error over 100 randomly partitioned datasets when 20 hidden neurons are selected WASHED OUT Training error Testing Error 4.2% 13.7% National Technical University of Athens, School of Chemical Engineering Interoperable Description of Cultural Material Content • Cultural material should be stored according to standardized formats • • Interoperability, Unified Accessibility, Portability, Exchangeability Metadata: Description of data • Extended Markup Language Schemes (XML’s) • MPEG-7 (visual description schemes) • MPEG-21 (resources and rights descriptions schemes) • Database schemas (MySQL, Oracle) • Tools for efficient Knowledge Search and Content Mining • Guidelines for new research efforts at European level • Great research effort, data alignment schemes, knowledge representation tools National Technical University of Athens, School of Chemical Engineering Conclusions-Guidelines •Artificial Intelligence can support automatic decisions for • Cultural material consolidation during the design phase • Finding the degree of importance for criteria used •Supervised learning => Feedforward Neural Networks •Unsupervised learning => Fuzzy k-means •Linear Regression (UTA *) => degree of importance •Results on Cleaning •Validation Performances •Issues on knowledge systems for cultural material National Technical University of Athens, School of Chemical Engineering Thank for Your Attention! National Technical University of Athens, School of Chemical Engineering