Interpreting Infrared Thermal Images of Repair Sites
FAA Sponsored Project Information A. Emery, K. Johnson, UW E. Casterline, J. Lombard, Heatcon Curtis Davies, David Westlund, FAA Technical Monitors Heatcon and Boeing 4/28/2011 2
Inverse/Optimal Repair of Composites Goal: To specify the spatial distribution of heat flux from a heating source to produce a specified and constant temperature throughout the cure zone with a minimum of pre-repair testing. Approach: Estimation of convective heat transfer coefficients using inverse numerical methods and Proper Orthogonal Decomposition of thermograms. 4/28/2011 3
Finite Element modelling The heat losses, Q, can be characterized by a surface heat transfer coefficient, h, Q=hA T h t h u h s h b 4/28/2011 4
Estimation of h s A 27 in. square panel was heated with a thermal blanket under various conditions while temperatures were recorded with thermocouples (TC s) and thermograms (TG s) A finite element model of the experiment was developed using COMSOL Multiphysics which outputs Temperatures. The h values in the finite element model were changed using standard inverse numerical methods until the least squares difference between the measured experimental temperatures and the simulated temperatures was a minimum. 4/28/2011 5
Thermocouple Placement and scan lines Surface TC Thermogram line scans Vacuum sniffer 4/28/2011 6
Measurements Thermogram scan In Between Temperatures are measured by thermocouples embedded in the composite, placed between the heating Blanket and the composite, and from infrared thermograms embedded 4/28/2011 7
Adding a Heat Sink to augment the heat loss Heat loss A channel with air flowing at different rates was used to mimic heat loss to structural members air flow 4/28/2011 8
Experimental Setup 4/28/2011 9
Program Outline 4/28/2011 10
Effect of Cooling Air No air flow 6.9 Estimated h b (w/m^2 C) 8.5 50 l/m 10.5 100 l/m 11.9 Blanket heating density= 2 w/in^2 14.1* 0.8 w/in^2 15.9 1.6 w/in^2 From measured Air flow and air temperatures * Common colors denote the results from the same tests 4/28/2011 11
Sensitivity of the Inverse Estimation 0 d T/dh 0.5 1 1.5 2 2.5 3 3.5 TC 9 ht TC 10 ht TC 9 hb TC 10 hb TG 9 ht TG 10 ht TG 9 hb TG 10 hb TC = thermocouple TG = thermogram 9 = over stringer 10 = over thin section 0 500 1000 1500 2000 2500 4 time(sec) 4/28/2011 12
Importance of Panel Properties The finite element model is highly dependent on the geometry and the material properties of the panel; i.e., thermal conductivity, specific heat, density. Literature values for properties of composite materials vary dramatically and are themselves highly dependent on volume fraction, resin type. Without accurate geometric measurements and material property values, estimation of h values is very difficult. 4/28/2011 13
Proper Orthogonal Decomposition M(t) M M N (x, y) Data N BP (Basic X M (Vectors) Pattern Vectors) Each column of is associated with a specific value of t and will recreate that data vector if is linearly interpolated between t1 and t2, it will produce the data vector that would have been observed at time t if the response were linearly related 4/28/2011 14
Information Content Each basic pattern vector is orthogonal to all others and the information associated with each vector diminishes as we progress from the 1 st to the last pattern vector 4/28/2011 15
The Stanford Bunny Let the M data vectors be the N pixel values of each column in the scan and get the basic vectors and see how many are needed to recreate the image? 4/28/2011 16
Recreation of the Bunny Of the 800 vectors, how many does it take to recreate the Bunny? diffe erences in Pixel Values 200 180 160 140 120 100 80 60 40 20 Stanford Bunny Pixel Values: maximum =238, std dev =65 Maximum Difference Average Difference Std Dev of Differences 0 50 100 150 200 250 300 Basic Vector # 4/28/2011 17
Energy Content of the Basic Vectors There are a total of 800 vectors, each 756 elements high 10 14 Energy Content of Basic Vectors 1.005 Energy Content of Basic Vectors for the Stanford Bunny Energy 10 13 10 12 10 11 10 10 10 9 Energy 1 0.995 0.99 0.985 10 8 0 1 2 3 4 5 6 7 8 9 10 Basic Vector 0.98 10 20 30 40 50 basic vector 4/28/2011 18
Typical Thermogram Scans 155 150 145 T (C) 140 135 130 No Channel Channel, No air flow 125 50 l/m 100 l/m 120-10 -8-6 -4-2 0 2 4 6 8 10 x 4/28/2011 19
The Basic Pattern Vectors 180 Y=23, basis vector 1 170 160 All look (C) T 150 140 no sink, no air 130 sink, no air 50 l/min 100 l/min 120-10 -5 0 5 10 x alike 4/28/2011 20
But Basic Vectors 3 and 4 show fundamental differences 6 4 2 Y=23, basis vector 3 no sink, no air sink, no air 50 l/min 100 l/min 2 1 0 Y=23, basis vector 4 T (C) 0-2 -4 T (C) -1-2 -3-4 no sink, no air sink, no air 50 l/min 100 l/min -6-10 -5 0 5 10 x -5-10 -5 0 5 10 x 4/28/2011 21
Conclusions The ability to accurately estimate heat transfer coefficients via standard inverse numerical techniques is highly dependent on the accuracy of the geometry and the material properties used in the finite element model. The basic (i.e., fundamental) patterns of the surface temperature as measured by the thermograms show promise in detecting backside heat flows. We anticipate that we can develop a library of basic patterns that are characteristic of different types of heat loss. What remains is to see if the higher order patterns can be used to quantify the magnitude of the heat losses and thus the needed blanket heating densities. 4/28/2011 22
Next Step Run tests using different levels of insulation and catalog basic pattern vectors. Insulation Determine if correlation can be drawn between basic patterns and heat losses. If successful this method would avoid all the difficulties involved with the inverse numerical approach i.e., required accurate geometry and material properties, ill conditioned nature of the problem. 4/28/2011 23
Questions? 4/28/2011 24