Anna Brook1 and Eyal Ben-Dor2 Remote Sensing and GIS Laboratory
Transcription
Anna Brook1 and Eyal Ben-Dor2 Remote Sensing and GIS Laboratory
Anna Brook1 and Eyal Ben-Dor2 Remote Sensing and GIS Laboratory 1The Porter School of Environmental studies 2Department of Geography and Human Environment Tel-Aviv University Modern high performance concrete - HPC The HPC is characterized by superior tensile properties and enhanced durability against severe environmental conditions offering a versatile and economic solution to many construction projects 2 Modern high performance concrete - HPC • Hydration - Term that describes a range of reactions between cement and water to produce the required strength • Curing - The process of maximizing the hydration of the cementations binder to achieve the intended design parameters • Hardening - Chemical admixtures are the ingredients in concrete to modify the properties of hardened concrete, to ensure the quality of concrete during mixing, transporting, placing, and curing, and to overcome certain emergencies during concrete operations 3 Current Concrete Analysis Methods • HPC are well characterized by their composition, documented in 'test samples' report from laboratory specimens and trial castings 4 The Problem In order to resolve specific problems that concern basic global urban environmental indicators, a new approach, including methods for in situ near real time analysis, is required! 5 Concrete Lab Test Concrete Cylinders, and Grout Cubes, are tested for strength according to the requirements of the WSDOT Standard Specifications. Concrete is also mixed in this lab and tested for Slump, Yield, Temperature, and Air Content 6 The Solution 1) Diffuse reflectance spectroscopy 2) The hyperspectral remote sensing Tools to assess the status and the strength of the concrete in situ 7 8 Spectral Models Spectral Database Standardized and normalized by an internal standard 3700 spectra Spectral Preprocessing Input spectral data Test Set Prediction Calibration set Logistic Regression Neural Networks Nonlinear iterative partial least squares Prediction ability evaluation Spectral Model 9 Logistic Regression • Selecting and Preparing the Input Dataset: 0.5 Log(1/R’) /R’) – spectral resampling – features selection algorithms 0.45 – spectral indices that produces Data for training subset • 0.4Statistic tests and models (M_ANOVA) 0.35 According to the spectral data we defined specific spectral regions where high and low spectral 0.25 variations occur using a1400 PC analysis, and spectral 400 600 800 1000 1200 1600 1800 2000 2200 2400 index in VIS (nm) region Wavelength 0.3 10 Logistic Regression Data Set 1 Convert reflectance spectrum (R’) to Log(1/R’) Data Set 2 Data Set 3 Physical/Chemical Properties Spectral Indices Multiple Analyses of Variance (M_ANOVA) PCA Spectrally Active Regions Partial Least Square Regression (PLS) Multinomial Logistic Regression Hydration Model Curing Model Hardening Model 11 Logistic Regression Hydration Model Curing Model Hardening Model Physical property mixture water-cement (w/c) and sand-cement (s/c) ratios Hardening: PVAA (Celvol 0-0.8% Curing period: 0,and 3, 5 and805) 7components: days at an early age Training 1000 Trainingdataset: dataset: 350spectra 850spectra spectra Testing 800 spectra Testingdataset: dataset: 300spectra 350 spectra R2 = 0.9582 Variable DF DFParameter Parameter Standard Wald Pr >Pr > Variable Standard Wald Variable DF Parameter Standard Wald Estimate Error ChiChi-ChiEstimate Error ChiEstimate Error Square ChiSquare Square 0.0001 Square INTERCPT 1 2.4563 0.3942 12.4245 Square INTERCPT 1 1 15.9435 0.4563 9.47320 0.0015 0.0115 10.12853 INTERCPT VIS slope 1-0.9452 -4.2195 0.5984 0.2104 10.0321 VIS slope 1 1.68743 0.1298 11.7637 0.0015 Iron oxides 1 -1.0839 0.4738 11.38477 0.0011 Hardener in SWIR 1 11.99087.215690.53828 0.4932 7.40134 4.372719 Liquid Clay 1 1 2.30556 0.4982 13.0466 0.001 0.0001 region water Hardener in SWIR 2 1 12.8642 0.1023 8.94315 No Change region <0. 5% Hygroscopic water (class) 1 0.36021 0.0022 13.799 0. 5-1% 1-2% Standardized Odds Standardize Odds Pr > Standardized Estimate Ratio d Estimate Ratio Chi- -Square 0.46397 0.0015 0.3927 0.39483 0.0001 0.47932 0.2053 Estimate 4.957 2.043 5.477 0.1932 3.875 7.309 Odds Ratio 12.92 0.001 0.0932 13.92 0.0015 0.1127 4.32 12 Logistic Regression – in situ 1m Hardening (%) 0.23 Progressive 0.25 0.25 Middle 0.23 Middle 0.23 0.23 0.5 m Hydration stage; w/c Hydration stage; w/c Hardening (%) Progressive 0.25 0.25 Early 0.21 0.2 Early 0.2 0.21 Aggregates and Sand in kg Coarse (1-2 cm) 100 Water/Cement Ratio Free Air in Balk (%) Hardener (%) 0.3 2.3 0.23 Fine (100Sand 500 mm) (200 µm) 50 90 13 Neural Network Input nodes Hidden nodes Training (correct rate %) Test (correct rate %) Output nodes First 8 PCs 3 100 100 3 First 5 PCs 3 99 95.7 3 First 4 PCs 3 98.6 95.1 3 First 4 PCs 2 82.9 79.5 3 14 Strength - Nonlinear Iterative Partial Least Square (NIPALS) Model Compressive strength (in MPa) N (concrete samples) 60 Mean strength (MPa) STD RPD 8 21.4 2.1 4.19 7 36.6 1.8 3.89 8 47.1 2 4.56 9 54.5 1.6 4.77 9 67.3 1.1 4.59 9 71.2 1.3 4.68 The laboratory results 55 50 45 40 35 Casting concrete day 21 - 352 spectra 30 5 10 15 Sample1 20 Sample 2 25 30 Sample 3 15 Hyperspectral Imagery 16 Classification Algorithm Spectral angle mapper (SAM) Mixture tuned match filtering (MTMF) Maximum-likelihood (ML) Minimum distance (MN) Neural networks (NNs) SAM Support vector machine (SVM) NNs WvA Wavelet analysis (WvA) Input HS image PPI Class “Concrete” Linear unmixing analysis Pure “Integrated” PPI Pure “Hydration” “Hydration” Spectrum SAM (with threshold) Possible Unmixing >0.1 DR SVM MTMF Binary MN ML Mask - 0 Binary Mask - 1 SAM MTMF ML MN LR model 2 NNs P(Hydration) SVM WvA 0.95 0.47 0.27 0.26 0.77 0.62 0.59 Add to Binary 0.63 - 1 0.47 Mask 0.49 0.91 0.86 0.88 Add to Binary Area - 0 0.97 Mask <0.1 17 Classification Algorithm Orange – hydrated concrete (prefabricated concrete walls) Magenta – early stage of hydration (crossbeams and windows frames) 18 Quality Assessment Aggregates & Sand (in kg) Coarse (1-2 cm) 90 Fine (100-500 mm) 40 Sand (200 µm) 80 Fine/Coarse Ratio 0.5 Water/Cement Ratio Free Air in Balk (%) Hardener (%) 0.3 1.9 0.3 Yellow- 0.3% w/c, Blue - 0.28% w/c, Green - 0.25% w/c 19 Case Study – in situ (21 days after casting) Aggregates & Sand (in kg) Coarse (1-2 cm) 100 Fine (100-500 mm) 50 Sand (200 µm) 90 Fine/Coarse Ratio 0.5 Water/Cement Ratio Free Air in Balk (%) Hardener (%) 0.3 2.3 0.23 Red is a–progressive stage ofofhydration - 0.25% w/c walls Green hydrated concrete prefabricated concrete Blue early stage of hydration - 0.2% w/c and windows frames Red is– an early stage of hydration of crossbeams Green is prefabricated concrete walls 20 21