A review of publications regarding detection of non-metallic inclusions in metal alloys using optical emission spectrometry with single-spark spectrum registration is presented. The main advantage of the method - an extremely short time of measurement (~1 min) – makes it useful for the purposes of direct production control. A spark-induced impact on a non-metallic inclusion results in a sharp increase (flashes) in the intensities of spectral lines of the elements that comprise the inclusion because their content in the metal matrix is usually rather small. The intensity distribution of the spectral line of the element obtained from several thousand of single-spark spectra consists of two parts: i) the Gaussian function corresponding to the content of the element in a dissolved form, and ii) an asymmetric additive in the region of high intensity values ??attributed to inclusions. Their quantitative determination is based on the assumption that the intensity of the spectral line in the single-spark spectrum is proportional to the content of the element in the matter ablated by the spark. Thus, according to the calibration dependence constructed using samples with a certified total element content, it is possible not only to determine the proportions of the dissolved and undissolved element, but also the dimensions of the individual inclusions. However, determination of the sizes is limited to a range of 1 – 20 µm. Moreover, only Al-containing inclusions can be determined quantitatively nowadays. Difficulties occur both with elements hardly dissolved in steels (O, Ca, Mg, S), and with the elements which exhibit rather high content in the dissolved form (Si, Mn). It is also still impossible to determine carbides and nitrides in steels using C and N lines. The use of time-resolved spectrometry can reduce the detection limits for inclusions containing Si and, possibly, Mn. The use of the internal standard in determination of the inclusions can also lower the detection limits, but may distort the results. Substitution of photomultipliers by solid-state linear radiation detectors provided development of more reliable internal standard, based on the background value in the vicinity of the spectral line. Verification of the results is difficult in the lack of standard samples of composition of the inclusions. Future studies can expand the range of inclusions to be determined by this method.
Determination of target detection limits in hyperspectral data using band selection and dimensionality reduction