In-process automatic wavelength calibration for CCD-spectrometers J. Mirapeix, A. Cobo, A.M. Cubillas, O.M. Conde, J.M. Lopez-Higuera Photonics Engineering Group, Univ. de Cantabria, Avda. Los Castros s/n, 39005 Santander, Spain ABSTRACT In CCD-spectrometers, the relation between the CCD-pixel number and the associated wavelength is established by means of a calibration polynomial, whose coefficients are typically obtained using a calibration lamp with known emission line wavelengths and a regression procedure. A recalculation of this polynomial has to be performed periodically, as the pixel number versus wavelength relation can change with ambient temperature variations or modifications in the optics attached to the spectrometer connector. Given that this calibration procedure has to be performed off-line, it implies a disturbance for industrial scenarios, where the monitoring setup must be altered. In this paper an automatic wavelength calibration procedure for CCD-spectrometers is proposed. It is based on a processing scheme designed for the in-process estimation of the plasma electronic temperature, where several plasma emission lines are identified for each spectral capture. This identification stage involves the determination, by means of a sub-pixel algorithm, of the central wavelength of those lines, thus allowing an on-line wavelength calibration for each single acquired spectrum. The proposed technique will be demonstrated by means of several experimental arc-welding tests. Keywords: arc-welding, on-line monitoring, plasma optical spectroscopy, CCD-spectrometer, wavelength calibration 1. INTRODUCTION CCD spectrometers are used in a wide range of different spectroscopic techniques, some of them devoted to the inspection of industrial processes, like food, tobacco or weld quality monitoring, among others. Both compact dimensions and low cost are important features of these devices. However, when the requirements in terms of optical resolution are significant, periodic calibrations have to be performed to avoid ambiguous or unexpected results. A good example can be found in the monitoring of welding processes, where on-line inspection is carried out in an attempt to identify the appearance of weld defects. Plasma optical spectroscopy has proved to be a suitable approach, as there exists a correlation between some spectroscopic parameters and the quality of the seams [1,2]. The correct identification of the atomic emission lines participating in the plasma is a key stage in this process, which will obviously depend on a correct spectrometer calibration. It is common practice to perform an initial wavelength calibration by the manufacturer of the spectrometer, where the pixel number of the CCD is associated with the corresponding wavelength or spectral band. This task can be fulfilled by means of a calibration lamp, which is a light source providing emission lines of known central wavelengths. In this regard, by performing a regression procedure, a calibration polynomial is obtained. However, this polynomial needs to be recalculated as the pixel number versus wavelength relation exhibits dependencies on the ambient temperature or the input optics attached to the spectrometer. If a calibration lamp or any other light source is needed for the calibration process, it implies that this process will most likely have to be carried out off-line. When an industrial monitoring system is considered, this calibration procedure could be an issue in terms of the process productivity, as it will force to stop at least the monitoring system itself. All in all, it seems clear that it would be desirable to perform this calibration procedure while the process is being analyzed by the system, and without including any external device, thus saving the costs involved. In previous works, we analyzed how to improve the stability of the plasma electronic temperature T e profiles by involving in the processing scheme sub-pixel algorithms [3]. This approach allows a more accurate estimation of the central wavelengths of the chosen plasma emission lines, which is the key parameter to perform an unambiguous identification of them. To identify a specific plasma emission line means to associate to it its corresponding chemical species and ionization stage. This process is typically carried out designing a local database with the spectroscopic information of the species participating in the plasma, which can be obtained from the NIST atomic spectra database [4]. Figure 1 shows the processing blocks considered for the estimation of T e in [3] (and the additional block devoted to Optical Sensors 2008, edited by Francis Berghmans, Anna Grazia Mignani, Antonello Cutolo, Patrick P. Meyrueis, Thomas P. Pearsall, Proc. of SPIE Vol. 7003, 70031T, (2008) · 0277-786X/08/$18 · doi: 10.1117/12.781040 Proc. of SPIE Vol. 7003 70031T-1 2008 SPIE Digital Library -- Subscriber Archive Copy