ABSTRACT
Composite materials are finding increased use in applications where impact and high strain rate loading form a significant part of a component’s service loads. It is therefore imperative to fully characterise the thermomechanical response of composite materials at high strain rates. The work described in the paper forms part of a project investigating the thermomechanical response of composite materials at high strain rates. To obtain the temperature evolutions during the high strain rate event (thermoelastic, viscoelastic and fracture energy), full-field infrared thermography is used. In contrast to visible light photography, the measurand in thermography is the intensity of the emitted radiation from the specimen surface, as opposed to reflected radiation. At increasing recording rates, the emittance available for measurement reduces proportional to the exposure time; the faster the data capture the less the exposure time. Hence, signal noise and detector calibration present a major challenge. This is accompanied by challenges arising from controlling an infrared detector that has not been optimised for the purpose of high speed data acquisition. The present paper investigates the possibility of applying infra-red thermography to high strain rate events and discusses the challenges in obtaining reliable values of the temperature changes that occur over very short time scales during high strain rate events.
KEYWORDS: Infrared, thermography, calibration, high speed testing
INTRODUCTION
High strain rate events, such as those that occur during a collision or impact, are known to invoke a different material response compared with that observed in quasi-static conditions [1]. One of the challenges associated with high strain rate testing is the short duration of the test and the required high data recording frequency. Here an infrared (IR) detector is used to obtain the temperature evolution of a specimen during high strain rate loading. The overall goal is to include the temperature data in constitutive laws that characterise the material behaviour. The present paper describes initial work on the implementation of high speed IR thermography in high strain rate tests using a commercially available system. The paper discusses in detail the challenges associated with obtaining temperature values from an IR detector output at high recording rates.
IR detectors are used in a wide variety of applications to measure temperature; the basic physics is described by Planck’s law [2]. At ambient temperature, the radiation band with the strongest emission is the middle IR band (1.5 – 20 ^m). In this range two types of detector can be used: bolometers where the detector experiences a rise in temperature due to incident radiation and a temperature sensitive material property (e.g. resistance) is measured or photon detectors where a semiconductor is excited by an incident photon and releases a charge that is collected in a capacitor. Of the two, photon detectors have a greater sensitivity and faster response rates and are therefore most suited to high speed data capture. To obtain a quantitative temperature measurement it is necessary to characterise the detector response. Calibration procedures are well established in industry and detectors that can be used to obtain quantitative measures of temperature are supplied with a manufacturer calibration. However, no commercial off-the-shelf system is available with a suitable calibration to obtain measurements at the recording frequencies required for high strain rate testing. The equipment used in the this work is a Cedip Silver 480M IR camera with a 320 x 256 element indium / antimonide (InSb) detector array. This is a photon detector, sensitive to radiation with wavelengths from 3 to 5 ^m. In standard operation, the detector has a sensitivity of 4.2 mK at 25°C, with a maximum frame rate of 383 Hz at full frame. To capture the temperature evolutions during a high strain rate test however, acquisition frequencies in excess of 10 kHz are required. Therefore the paper describes in detail the characterisation and calibration of an off-the-shelf detector for high speed thermography.
DETECTOR CHARACTERISATION
Fig. 1 shows a schematic of the data capture for a photon detector such as that used in the 480M system. The photons are focused on the detector array. The electronic shutter controls the exposure time, known as integration time (IT), by controlling the time the switches are left open. To achieve the highest possible frame rate and sensitivity, the system comprises two sets of capacitors in the read out circuit, enabling almost continuous data capture; while data is being read from one bank of capacitors, the second bank is recording. The output from the capacitors is converted in to a 14 bit logic signal using analogue to digital converters built into the detector device. The digital output is then sent directly to the computer for further processing.
Fig. 1 Schematic of data acquisition
Selecting a suitable frame rate is a compromise between three main parameters: the duration of the test, the required thermal sensitivity and the number of detector elements used to collect the data. The frame rate is limited by the (IT), which controls the detector thermal sensitivity and the data handling capacity of the read out circuit which dictates the maximum window size for the image (i.e. number of detector elements) that can be recorded at a given frame rate. Fig. 2 illustrates how the three parameters are related for the system used in this work and shows that a practical limit is reached around 16 kHz above which the sensitivity and image window size fall below a useable threshold.
Fig. 2 Relationship between a) integration and frame rate and b) window size and frame rate
Each detector element has a slightly different sensitivity and similarly each capacitor has its individual characteristics. To obtain quantitative measurements, the first step is to identify this variation so as to be able to correct for it. This process is called the non-uniformity correction (NUC), which comprises exposing the detector to a uniform diffuse emission source (generally a flat plat with a high emissivity coating) and recording several images. The average signal from the whole image (spatial average) is compared with the average signal of each detector element (temporal average). This is done at two different temperatures within the range of temperatures to be measured, as shown in Fig. 3. A linear fit is then assumed between the two points, providing a gain and an offset value for each detector element from which the correction factor can be calculated, to bring it in line with the mean response of the whole array. This linear fit is however only an approximation of the detector response curve. In a typical test setup, the IT is selected to maximise the use of the full range of the detector (typically from 25 to 75% of the detector saturation). The two points for the NUC are chosen at 30 and 70% of the detector saturation to minimise the average error across the full range. The NUC is stored in two tables (one for each array of capacitors) which are loaded directly into the flash memory onboard the camera so that the data output by the camera is already corrected. In the current work however, the IT time is selected to enable a particular frame rate to be achieved. At this IT the detector will only operate between 2 and 10% saturation for room temperature measurements, so the NUC was performed at 30 and 70% of the desired temperature range (i.e. 28 and 51°C for the range 10 to 70°C). Over such a small range, the linear fit assumed in the NUC should be a relatively good approximation.
Fig. 3 Linear fit for NUC
CALIBRATION
As the necessary IT is so small, the calibration data produced by the manufacturer does not cover this range since it is well outside the usual operating conditions for the IR system. It was therefore necessary to devise a methodology to establish a calibration at small IT. The methodology followed that of the manufacturers approach by taking the detector output in digital level from a high emissivity body with a known temperature. In this work a cavity black body was used [3] with a platinum type thermocouple inserted. The cavity walls were maintained at a uniform temperature by water circulated from a temperature controlled bath with a temperature range from 4 to 75°C. The calibration was conducted twice; the first time a series of videos (20 images each) was collected over a temperature range from 8 – 50°C using an IT of 60 ^s and a frame rate of 15 kHz and an image size of 64 x 12 pixels. A NUC was conducted at 10 and 30°C. A calibration curve of detector response against black body temperature was obtained for odd and even numbered frames separately to compare differences that might occur due to the two capacitor arrays. The results showed an almost identical response. It was also possible to obtain the noise in each detector element, calculated as the standard deviation from 10 readings at nominally identical temperature. A standard deviation between 1.5 and 2.5 DL was obtained for most detector elements, with a few pixels having a standard deviation up to a maximum of 3.5 DL at 50°C. This represents a measurement precision of 0.35 to 0.2°C over the range from 15 to 50°C. The calibration procedure was then repeated over the range from 5 to 70°C, this time taking only single images at each temperature to reduce the data processing. A second NUC was performed, this time at 28 and 51°C to account for the extended temperature range. From this data the standard deviation across the image was obtained. This showed a nearly Gaussian distribution at temperatures up to 30°C. Fig. 4 shows that at 50°C, the pixel value distribution across the image shows two spikes, and a greatly increasing spread. Viewed as a percentage standard deviation the noise appears to be quite small, as shown in Fig. 5, but the change in the image noise characteristics seems to display a systematic pattern in the detector response and must therefore be considered significant. This is attributed to an error with the NUC as the pattern in the output has been shown to match the noise pattern of an image obtained with no NUC. The outcome is counter intuitive as the lowest error associated with the NUC process would be expected at the two points at which the NUC was conducted, i.e. at 28 and 51°C. However, it is at 50°C where two spikes start to appear in the image histogram as shown in Fig. 4. The signal noise from an individual pixel, however, remains fairly constant across the temperature rage, increasing from 1.5 to 2.5 DL between 15 and 70°C. Hence, the increase in noise is not a function of the detector elements themselves but of how the NUC process is integrated into the onboard hardware.
Fig. 4 Histogram of image noise at different temperatures
Fig. 5 Change in image standard deviation with temperature
The error with the standard NUC procedure can be avoided by obtaining raw detector data and performing the NUC in a postprocessing stage off-line. The strategy in this work is to conduct a calibration on a pixel-by-pixel basis. This takes into account the responsivity of each individual detector element. The calibration has to be conducted separately for odd and even frames to account for the two arrays of capacitors in the data acquisition hardware. By conducting the calibration directly on the individual pixel response, the two steps are combined. This enables the measurement precision to be assessed for each pixel individually in the resulting image. This calibration was conducted with 2°C intervals across the calibration range from 10 to 70°C. The plot shown in Fig. 6 shows the image average for odd and even frames. The error associated with each measurement is less than 4 DL. Table 1 shows the aggregate detector sensitivity (temperature increment per DL) and precision (detector noise calibrated into °C) over the measurement range.
Fig. 6 Calibration curve (image average for odd and even frames)
Table 1 Detector Precision
|
Temperature (°C) |
Sensitivity (°C) |
Precision (°C) |
|
15 |
0.14 |
0.35 |
|
20 |
0.11 |
0.27 |
|
30 |
0.10 |
0.25 |
|
50 |
0.07 |
0.20 |
The final step in obtaining accurate measurements is to evaluate the emissivity of the surface of the test specimen. For this, material coupons were placed in a thermal chamber with a temperature range up to 80°C and a small hole cut into one side to provide optical access to the IR detector. Thermocouples were mounted to the rear of the coupons while the forward facing side was prepared in the same way as for a mechanical test. Images were then collected over a range of temperatures from 15 to 55°C to assess the emissivity of the specimen surfaces over the full range of temperatures expected to occur during the high rate testing. Because the specimen surface is not a perfect emitter, background radiation will be partly reflected from the surface, leading to a measurement error. To assess the effect of a background radiation source, the specimens were also placed at a slight angle to the detector (approximately 20°) and a diffuse emission source (a flat plate with a high emissivity coating) was placed in the reflection path. The emissivity of lightly abraided E-glass / epoxy composite was measured to lie in the range 0.90 – 0.92. This uncertainty is due to some inherent noise in the detectors and variability in the surface preparation.
TEST RESULTS
Preliminary tensile tests were performed on an E-glass / epoxy specimen made from 4 layers of chopped strand mat to assess the range of temperatures to be expected from a composite specimen in the approach to failure and evaluate the calibration and NUC procedure described above. The specimen was waisted to provide a gauge length that fitted entirely within the field of view of the detector (approximately 10 x 50 mm). The mechanical load was applied using an Instron VHS servo-hydraulic test machine at an actuator velocity of 10 ms-1 and a maximum load capacity of approximately 30 kN. The load was measured using a Kistler piezo electric load cell. However the load data was extremely noisy and could therefore not be used. As a strain gauge was not attached to the specimen directly, the strain rate was estimated using the gauge length, actuator velocity and the assumptions of no slip in the grips and was considered to be between 50 – 100 s-1. The image in Fig. 7 shows the temperature on the specimen surface at the time of fracture. The fracture location is clearly visible on the right hand side of the specimen as a region of significant heating. Temperature measurements taken at five locations on the specimen surface (as marked in Fig. 7) are plotted over time in Fig. 8 a) and b) from 1 ms before fracture to 1 ms after. Fig. 8 b) shows a zoomed in portion of the graph in Fig. 8 a) enabling two distinct regions to be identified: an elastic strain region where there is a uniform decrease in the temperature across the whole specimen due to the thermoelastic effect [4] and a region where the specimen temperature increases all across the specimen surface as the failure strain is reached. The final failure propagates from one location on the specimen (point 5). Here heat is generated by the formation of new surfaces. Post failure inspection indicated manifold failure modes, including fibre pull-out, fibre failure and matrix cracking at the fracture site.
Fig. 7 Image of specimen surface temperature at time of fracture
Fig. 8 Surface temperature evolution at 5 points on the specimen, a) full range b) zoomed in
CONCLUSIONS
It has been demonstrated that an off-shelf IR detector can be used to obtain temperature measurements at high speed. In doing so it is necessary to carefully investigate how the detector behaviour changes. Manufacturers’ calibration processes are often integrated into the detector hardware in a manner that may not be transparent and possibly even a trade secret. Apparently straight forward processes such as the NUC may therefore not perform as expected when operating outside the detectors normal range. The work in this paper has highlighted the procedures necessary to obtain quantitative temperature measurements as well as some of the potential pitfalls to be aware of. It is demonstrated that data can be obtained with sufficient precision to make a meaningful addition load and strain measurements during high speed testing.
