Digital Industrial Radiography Dr. Helmut Wolf, Anna University Chennai Dr.Theobald Fuchs, Fraunhofer Development Center X-ray Technology, Fuerth, Germany 1. Introduction In the previous issues, the physical principals of industrial radiography (RT) and high resolution radiography were discussed. Digital radiography (DIR) is the topic of this article. DIR is based on the same principles of imaging as RT, i.e. the recording of spatially resolved radiation intensities. The only difference is that the method of recording and visualization follows a digital route. We always have to keep in mind that we deal with X-ray absorption images and not optically focused photographs. Digital images are nowadays found in many applications. In photography they almost completely replaced film imaging. In RT, the process of transition to digital methods is slower, mainly for economic considerations. 2. Fundamentals 2.1. Analog and digital When discussing images, often times the distinction between analog and digital is made. Analog means proportional or continuous. Digital means discrete, or stepwise. Computers require discrete numerical values for processing and storing. The conversion of an analog signal or intensity to a number means that a fixed step value is assigned to a small continuous range. The process is called A/D conversion or digitization. The quality of digital conversion depends on the number of steps we assign to a continuous range of values. 1-D signal position Figure: Converting an analog range to digital values. Each signal is assigned to the closest discrete value out of a finite number of values at typically equidistantly sampled positions. In the case of radiographs, a film is often considered analog. This is it not quite correct. The darkening of the film after exposure and development is due to very tiny silver crystals, also called the film grain. It is only because the eye cannot normally resolve the individual grains that a radiograph is considered continuous. Actually these grains are randomly arranged, separate particles, a few micrometers in size. When viewing a film the grains can normally not be seen individually, but are perceived as a continuous variation in density. 2.2. Digital images Digital images are made up of a fixed rectangular arrangement of square 1
dots. The requirement of storage of a digital image can be very large depending on the parameters of a digital image which are discussed below. Spatial resolution. The dots that make up a digital image are called picture elements or pixels. A pixel has a definite size. In high resolution images a pixel may have a size of 50 µm x 50 µm. Actually, this size is about the best resolution the human eye can achieve and again, when we view such an image, we would think it is continuous in space and not broken down to individual elements. If we take a standard 5 cm by 20 cm radiograph at 50 µm pixel size (i.e. 200 Pixel per cm), this image requires 5 x 200 x 20 x 200 or 44 million pixels. Depth resolution. One more property of a digital image is important. We can assign intensities or shades of grey to each pixel. If only black and white is required, we can represent this in terms of 0 or 1. We would describe it as every pixel having a depth of one bit (2**1). More shades can be represented with a larger pixel depth. If we take an 8 bit pixel depth, we have 2**8 = 256 shades. Mostly, instead of bit, the unit byte is used, 8 bit = 1 byte. In this case one pixel has the storage requirement of one byte. Again it is interesting to take the human vision as a reference. We can distinguish two shades of grey, if their difference is at least 2%. This is also called visual contrast resolution. For the whole range of greys from black to white, at best we can recognize only 50 separate shades, each 2% apart. This means that a digital image that is based on 8 bit resolution appears as a continuous image to the human eye. However, a digital system is capable to meaningfully resolve and record as many as 12 bit (4096 shades) or 16 bit (65536 shades). This also increases the storage requirement. If we take the radiograph of the above example and assign 12 bit (= 1.5 byte) for every pixel, 44 million pixels require 66 million bytes or roughly 63 MB of storage for a single 5 cm x 20 cm radiograph. Temporal resolution. For completeness, we look at the resolution in time, which could be important for real time radiography. Movies take advantage of the fact that we cannot resolve individual images, if we see more than about 25 frames per second. A movie appears to be continuous, though objectively, we are presented with a sequence of still photographs. Considering these fundamentals of digital imaging, we can understand how digital radiographs can contain more information than what can be seen with the naked eye. They can be superior with respect to all resolutions - space, depth or time. All we need is sufficient memory space and accurate conversion devices. In this way we can obtain more information that is immediately visible. Digital images can be numerically evaluated or visualised by digital processing. The image can be magnified (spatial, also called pixel mapping), a limited tone range can be converted to a larger 2
range of shades (depth mapping, contrast adjustment) or the time intervals of display can be increased (time mapping, slow motion). Digital images are also required for automated systems where features are automatically detected and evaluated. In Computed Tomography (CT) images are superimposed for reconstruction of the inspected volume. Without digital technology CT would not be possible. 3. Creation of digital radiographs There are different procedures to arrive at a digital radiograph. Conventional film can be a starting point. Real time systems often use conversion by scintilization (light flashes) combined with photomultipliers. Computed radiography uses storage phosphors that are read out after exposure. Only recently directly converting flat panel detectors have become available. 3.1 Conversion of conventional radiographic film Conventional radiographic film is converted into digital images for a number of reasons. An important reason is that radiographic film contains more information than can be seen, as discussed earlier. We can detect more details in an image. When ISNT held a first DIR workshop in India in 1999, the cost of storage media (DVDs, Hard Disks etc) was so high, that it was not economical to keep scanned image files for storage purposes. However, the advantages of processing and evaluation were sufficient reason to digitize films. A number of digitization procedures were developed: Point scanners. The digitization process reads the film density and converts it to a numerical value. The first digitizers worked with a single light source one side and a light sensor on the opposite. The principle is the same as in a film densitometer, except that the density is recorded electronically and within smaller areas. To read an entire film, the film is moved point by point by a mechanical X-Y scanner and the pixels with the density values are assembled to form a digital image. Line scanners became available, where an image is read line by line. The scanner consists of many sensors along one line. The sensor is moved by an index value and the pixels covering a complete film area are assembled into a digital image. This procedure is very much like in an office scanner, except that office scanners work in reflection. X-ray film scanners work in transmission mode and need the ability to process larger density ranges, as much as 1 to 10000, corresponding to densities of 4. 2-D scanners. The film can be read in a whole area. The pixels are defined by the resolution of the optical sensor. These are often CCD cameras. Since CCD sensors have a limited dynamic range, X-ray images are often processed with multiple exposures, each exposure in a different dynamic 3
range, to preserve the maximum density nuances of a film image. cone-beam hidden internal detail 3.2 Computed Radiography Computed radiography (CR) is a method that uses imaging phosphor plates. The radiation received is stored in the phosphor and read out by thermoluminescence effect. The work flow is very similar to conventional radiography. The plates can be handled like film, even bent around a weld. In principle, the plates can be reused more than 1000 times, but in practical applications this is hardly reached, because any mechanical damage or finger print shows up on subsequent images. CR is ideal for laboratory environments where the plates are not handled, but placed in cassettes and automatically processed in reader scanners. A laser beam is stimulating visible light emission proportional to the radiation exposure of the plate. In a special read out scanner, the laser beam is focused on one spot. The laser stimulates the emission of light at one spot. The spot is shifted by a rotating prism and covering the entire area of a film. This readout process can only be performed once. By reading the plate, the latent image is removed. The process cannot be repeated. The parameters of the digital image, especially the pixel size is dependent on the focal spot of the laser. Each spots becomes one pixel of the digital image. point source sample object s shadow flat panel detector or digital detector array (DDA) or 2-D X-ray sensor or X-ray matrix detector Figure: Projection imaging geometry 3.3 Digital Flatpanel Detector Arrays - What is a flat panel detector array? A Flat-panels detector array (FDA) is subdivided into pixels already. Each channel of the flat panel matrix can be considered a separate X-ray detector, comprising - a photo-diode/capacitor - a TFT switch for read-out, - followed by an amplifier, a multiplexer and an analogue-todigital converter (ADC). There is a strong similarity to an instrument in nuclear physics: A large number of individual detector channels (10 5 up to 10 7 ) are assembled within the same electronic device the detector matrix. row line Figure: Arrangement and Internal structure of a typical FDA data line bias line photo diode TFT switch one pixel 4
Signal characteristics of real-world flat panel detector devices FDAs have to be understood as a complex electronic instrument for signal acquisition and processing: the signals contain thermal electrical noise; often there are bad channels (black or white pixels); there might occur some coupling with electromagnetic fields in the cables and the electronics; there is a finite digitization depth (number of bits of ADC); there is a read-out time (dead time) and a cycle time (given as frames per sec); the signal from each pixel of the matrix ( intensity in grey values) is a measure of the number of photons absorbed by that particular pixel during integration time. A FDA is usually not one single piece of amorphous silicon of 200 mm or 400 mm lateral size, but the detector matrix is made up of several tiles with a read-out chip assigned to each. Thus, the raw images reflect the internal structure of the device. The there can be overlaying patterns that are characteristic for a particular detector arrangements of detectors. Since quality of semi-conductor material and processing of the microelectronics may be inhomogeneous, the sensitivity of the channels may vary from chip to chip, from line to line, and in large irregularly shaped areas (clouds). If we want to obtain a uniform image output, for a uniform input, we have to calibrate every pixel electronically to compensate differences in linearity and sensitivity. read-out lines zoom Figure: Uncorrected dark image. The thermal noise caused by the read out electronics can be seen clearly (left hand side: full panel, right: zoom). Gain-offset-correction Each pixel has to be treated as an independent measurement channel. The electrical signal from an individual detector pixel can be written in a linear approximation: v ( I) v + I g = dark defect pixels Read-out lines A single read-out chip Dark cloud 2048 x 1536 pixel Thereby, I denotes the X-ray intensity reaching a single pixel. Each individual channel is characterized by its dark current v dark and gain g. These parameters are generally unknown and have to be determined by calibration measurements. The dark current (offset) is measured with zero X-ray dose: I = 0. The bright image is measured with the primary intensity I 0 applied during 5
measurement. Its calibrated value can be chosen arbitrarily, e.g. v calibrated (I 0 ) = 40.000 grey value level. Usually, the intensity (e.g. dose or photon flux) is not measured directly by an additional instrument. Thus, for reasons of practicability, the gain is not measured directly. The variation in gain between different channels is corrected by applying a scaling factor. v calibr. ( I) v = v ( I) ( I ) 0 v v dark dark v calibr. ( I ) 0 digital signal Max = 2 b - 1 ideal channel (calibrated) bright image intensity I dark image I 0 Figure: Digital output signal as a function of the X-ray intensity measured for every detector pixel. Each of the three lines stands for a particular detector pixel. Commercially available FDAs There are various types of flat panel detectors commercially available today. The devices offered by several manufacturers vary in pixel size, pixel format, the type of X-ray conversion, read out frequency and last but not least price. Figure: commercially available FDAs PerkinElmer (left), Hamamatsu (center), Vidisco (right). pixel size: 50 µm up to 400 µm area: 50 mm x 50 mm up to 400 mm x 400 mm read out cycle between 5 and 30 frames per second indirect conversion (scintillator CsI, Gd 2 O 2 S:Tb) 6
amorphous silicon (Perkin-Elmer, Varian, Trixell, GE) CMOS (Hamamatsu, Rad Icon) direct conversion Cadmium-Telluride (Ajat, MediPix) Gallium-Arsenide (MediPix) CCD-based Selenium The costs for a FDA range between 10.000 and 100.000 Euro today, but prices are decreasing as the numbers of sold systems increase. The life-time of a FDA is limited due to the inevitable radiation damages to the micro-electronics. According to the experience of the authors the life-time of a device which operated in a 24/7- mode is between 12 and 36 months. 4. Digital image processing Exactly the same way as in digital photography, nowadays, there are countless methods for digital image processing, which can be applied to the X-ray images. One of the most simple options is the enhancement of the images by use of local or global filters. A very common example is the median filter which helps to smoothen regions of the image where the dectability of details suffers from high image noise. As well median filters serve in removing irregular pixels while preserving edgelike structures. Another example of common tools for the enhancement of images are look-up tables which serve to adapt the grey scale range for visualization purposes. The same, contrast and brightness of digital X-ray images can be modified easily and repeatedly. Moreover, the digital technology allows for a computer aided analysis of any kind of X-ray images. Algorithms for pattern recognition and feature extraction can be applied, for instance in order to detect automatically voids, cracks or inclusions in various kinds of materials and components. Data fusion with other NDT methods is possible as well as more complex operations like the reconstruction of 3-D volume data sets from projections in Computed Tomography. A detailed treatment of the latter techniques is far beyond of the scope of this publication, but of course they and all other image processing methods require digital images as input data. 5. Automatic Defect Recognition (ADR) in industrial production As an example for a fully automatic inspection system, the Fraunhofer ISAR system is capable of acquiring a single digital radioscopy image within 200 milliseconds. Typically, for each part 3 to 14 different images are acquired, and evaluated, thereby keeping up with a production cycle of about 10 seconds per part or less. Subsequently, the software makes a decision, if the current part can be accepted as defect-free, according to the limits which are prescribed to the component manufacturer by the OEM. 7
6. Digital images, issues of archival, durability and integrity 6.1. Archival and durability Films deteriorate over time. Films that have not been washed well or are stored in humid places develop spots and patches soon, but there are well processed and stored films that have survived more than 100 years. As for digital images, it is true that there is no deterioration of images. The data can be copied any number of times and there is no loss of quality. However, if a set of data becomes unreadable, the loss is total. This can be due to mechanical damage of medium such as a hard disk or CD. If long term archiving is required, it is more likely that data become unavailable because the reading devices become technically outdated. Just consider diskettes or data tapes. Today you really have to search for readers to recover old data. 20 years from now CD and DVD drives may be hard to find.experts consider this to be the more important issue. Accidental loss of data can be taken care of by multiple copies, stored at different locations, but the loss of data due to obsolescence also has to be addressed. Figure: Fully automatic X-ray inspection system of aluminium castings. Casting and three digital radiographs processed for defect detection. 6.2. Data compression It is obious that the availability of digital X-ray images leads to a previously unknown amount of data. In particular, 3-D volume data afford a large amount of storage, a challenge that is to be addressed in medical imaging but as well as in industrial inspection. On the 8
other hand a manifold of algorithms for data compression is available today. In general, these algorithms can be devided into two classes: lossy and lossless compression techniques. The decision on which type of compression is to used has to be made case by case. Although the efficiency of lossy compression algorithms in terms of storage saving is higher than with lossless methods, there may be legal requirements or safety issues which prohibit any reduction of information within the digital X-ray images. Overall, the rapid progress in digital imaging has also affected the full spectrum of industrial X-ray inspection. A large variety of new methods and inspection possibilities are emerging with these new devices. 6.3. Data integrity X-rays are often required for legal purposes and digital images have the reputation is that they can be altered or manipulated easily. Common software allows to add or remove image details. This is also possible in digital X-rays. Today there is no universal standard or a fool proof system that can guarantee that there has not been an alteration of an image. In principle authentification systems are possible, but have not been standardized. 7. Summary This paper discusses the basic principles of digital X-ray imaging, further techniques like film digitization and computed radiography (CT). The latest development of flat panel arrays (FPA) is explained in detail with respect to possibilities and challenges. A number of applications of digital X- ray images are pointed out, as well as the issues of image processing, compression and archiving. 9