Towards a Universal Metals Identifier - A Compact CCD Array Spectrometer

Trevor Smith

The development of a compact optical emission spectrometer based on a linear CCD array detector and a holographic flat field diffraction grating is described. This instrument offers many features that are required of on-site metals identification applications. The system allows the selection of several hundred lines and estimates relevant element concentrations in metals and metal alloys of eight base elements (Aluminium, Copper, Zinc, Iron, Nickel, Titanium, Cobalt and Magnesium). In a single test the following is performed: the base element is identified; the spectrometer is automatically profiled to optimise channel location; an individual matrix (curve set) is identified automatically, where appropriate; element concentrations are estimated; and the material grade is identified from a list of the specifications used by the customer. Performance results for base identification, concentration estimation, accuracy and repeatability, and the correct identification of grade are reported.

Environmental awareness in industry, frequently driven by new legislation, has increased the demand for efficient metals recycling. Also, intensified competition as a result of world recession has increased the demand for quality control and hence for positive identification of metals. These two factors result in an increased need for efficient low cost metals identification equipment for use on site in the industrial environment.

The coverage of metal type should be comprehensive, and include all the common base elements: Al, Cu, Zn, Fe, Ni, Ti, Co, Mg, Pb, Sn and Zr. Such an instrument must be equally capable of sorting scrap material, positive material identification (PMI) and sorting out mix-ups of stock material.

X-ray fluorescence (XRF) and optical emission spectrometry (OES) are the only practical alternatives for these types of application and recent technical developments have concentrated on the latter. For cost reasons, several manufacturers have turned to solid-state detector array based spectrometers employing holographic flat field diffraction gratings. However, the comprehensive coverage defined above has so far only been an ideal. In conventional spectrometer designs, the sensitivity of photo multiplier tube (PMT) detectors varies between channels and over time. However, with the use of a semiconductor detector, these two effects are largely eliminated. However, every spectrometer has slightly different characteristics with regard to sensitivity and degree of focus at each wavelength and it is still necessary to calibrate each spectrometer for every type of material for which it is to be used. Moreover, calibration for a comprehensive material coverage becomes expensive because matrix interferences and non-linearities are rife in electrical plasmas, and these effects must be properly characterised in order to achieve reasonable accuracy for all materials.

A major weakness in the design of some portable metals spectrometers has been the use of a fibre optic coupling between the source and the spectrometer. In these instruments, the spectrometer has been made compact by the use of miniature PMTs, which are generally more expensive than their regular sized counterparts. The spectrometer is housed in the controller and coupled to the hand held source by a bulky cable containing the delicate optical fibre together with the electrical supply to the source and various control signals for buttons and indicator lights on the probe. Fibre optic coupling ensures that light from a wide cone angle is collected at the fibre end and this makes the optics less sensitive to movement of the arc or spark and the relative positions of the sample and the counter electrode. However, the efficiency of light coupling is drastically reduced (Webb, 1989).

A radical departure from this configuration is possible by the use of a miniature spectrometer that can be contained within the hand held probe. This allows the delicate fibre optic with its associated attenuation to be eliminated but obviously requires careful consideration of weight and robustness if the probe is to be employed continuously over, say, an 8-hour shift in a harsh industrial environment.

The spectrometer consists of a hand held probe containing the spectrometer, detector drive electronics, the d.c. arc source together with its associated spark ignition system, two control buttons and three indicator lights. The probe is connected to the mains powered controller via a light-weight electrical only cable. The controller contains an IBM PC compatible computer running an embedded application program under the DOS operating system. The facilities provided include a keyboard (built into the lid of the controller), a backlit high resolution LCD display, a floppy disk drive, a 16 column printer and the power supply for the d.c. arc. The arc power supply and all other aspects of operation of the instrument are fully computer controlled so that operation is simple and can be accomplished, for normal testing purposes, by three buttons on the front panel of the controller plus the two button on the probe. The keyboard and disk drive are only used in configuring and setting up the instrument. The basic optical configuration of the spectrometer is extremely simple (Fig. 1.)

A custom holographic diffraction grating of 153mm radius of curvature is positioned 170 mm from a 10µm wide slit but 1° off-axis. The d.c. arc source is approximately 25mm directly in front of the slit with no intervening optics to enhance light collection. This configuration allows the detector to be positioned slightly behind the slit leaving room for the source. The custom grating produces a focus field that deviates from flat over the length of the detector by only ±0.3mm (see Fig. 2.)

The detector is run with an exposure period of either 0.0625 or 0.125 sec per scan. Under normal burn conditions a 2.5A arc is struck by means of a spark ignition system and allowed to burn for about 1 sec to allow stabilisation. Scans are accumulated over a 4 sec period. Data are suitably scaled and truncated to 16 bit integers. Every 30 seconds or after each burn if burns are in-frequent, a dark current reference signal is accumulated over 1 sec and subtracted from each accumulated signal spectrum. This removes dark current, an offset introduced in the electronics to ensure that the signal at the analogue to digital converter is always within range and fixed pattern dark noise on the detector.

For manual selection of the calibration curve set, appropriate base element and analyte line intensities are estimated from the spectrum. These intensities are then used to estimate concentration of each element using calibration curves which are up to 3rd order polynomials. Results are 'normalised' to ensure that concentration estimates of all the elements add up to 100%.

When used in automatic mode, signal processing commences with the intensity and position estimation of key element lines used to identify the base element of the material. A second or subsequent layer of curve set selection can be used to identify particular types of material within one base element. Once the appropriate curve set (matrix) has been identified, processing is identical to the manually selected case.

An important feature of the line estimation algorithms used in this instrument is the automatic profiling capability. The positions of key base element lines are used to determine the precise location of the spectrum on the detector to within a fraction of a pixel width. The position of analyte lines is then estimated from a model of the pixel position vs. wavelength relationship (a 4th order polynomial is found to be accurate to within 0.05 pixel over the entire 185 to 410nm range). The automatic profiling is performed individually on every burn and can accommodate up to ±3 pixels of movement of the spectrum due to thermal expansion or other effects. The nominal temperature range for operation of the spectrometer is 0°C to 50°C.

A study was conducted over a one year period to identify the principal metal types which a metals identifier should be able to recognise. In this study, the capability to measure C, P or S was specifically excluded. As alloying constituents of carbon steels, these elements are obviously very important. However, it was observed that carbon steel was not as frequently requested as might be expected. This may be because, in terms of value, the need to identify or check grades of carbon steel is not so important.

Table 1 indicates the groups of metals identified. In some cases the metals groups have been combined for simplicity. For example, there are numerous groups within nickel base, but four principal groups have been identified: monels (high copper content); incoloys (high iron content); nickel-cobalts (high cobalt content); and then all other nickel base alloys grouped together. More detailed information was gathered, including which elements are important within each matrix. This data is not presented here for brevity.

 Certified calibration samples representing all the above materials (except Pb, Sn and Zr base) were used in the study. Typically, each matrix was represented by between 6 and 20 samples.

For experimental purposes, data were collected directly onto diskettes (3.5") using a special data capture program and all analysis, calibration and evaluation of results performed off-line on PC compatible computers. Each sample was burned 6 times at room temperature (nominally 25°C). Selected samples in the set for calibration of each matrix were also burnt 6 times at both 0°C and 50°C. This allowed checking that the calibration is valid at these temperatures and that the automatic profiling is functioning.

Eight bases and 16 matrices have successfully been calibrated (fig 3). Some optimization of burn conditions was achieved. It was found that reasonable results could be obtained with just two burn conditions (A and B). In many cases the final result can be achieved from the initial burn (condition A). For the lighter metals (Al, Zn and Mg base) the initial burn can be used to identify the base element but, to achieve a full approximate analysis of the alloy content, burn condition B is required and a second burn must be performed. There are in addition some other observations to be made: Almost all the copper base materials can be grouped together in one curve set: the copper alloy matrix. Silicon bronze, however, cannot reliably be separated from copper alloy automatically so the silicon bronze matrix, which allows better accuracy for high silicon copper base materials, must be selected manually. Similarly, Carbon steel, whilst it also uses the same burn conditions as other steels, is a curve set only accessed by manual selection; the default automatically selected matrix for a carbon steel is 'Low Alloy' steel.

Hopkins, G.W., Nordman, R.G., and Willis, B.G., 1979, Optical and opto-mechanical design of a high-throughput ultraviolet(UV)-visible spectrophotometer, SPIE 191, pp 48-55

Webb, M.J., 1989, Practical Consideration When Using Fibre Optics with Spectrometers, Spectroscopy, 4(6), pp 26-32