Enhancing the signal-to-noise ratio of X-ray diffraction profiles by smoothed principal component analysis

X-ray diffraction is one of the most widely applied methodologies for the in situ analysis of kinetic processes involving crystalline solids. However, due to its relatively high detection limit, it has only limited application in the context of crystallizations from liquids. Methods that can improve the detection limit of X-ray diffraction are therefore highly desirable. Signal processing approaches such as Savitzky-Golay, maximum likelihood, stochastic resonance, and wavelet transforms have been used previously to preprocess X-ray diffraction data. Since all these methods only utilize the frequency information contained in the single X-ray diffraction profile being processed to discriminate between the signals and the noise, they may not successfully identify very weak but important peaks especially when these weak signals are masked by severe noise. Smoothed principal component analysis (SPCA), which takes advantage of both the frequency information and the common variation within a set of profiles, is proposed as a methodology for the preprocessing of the X-ray diffraction data. Two X-ray diffraction data sets are used to demonstrate the effectiveness of the proposed approach. The first was obtained from mannitol-methanol suspensions, and the second data set was generated from slurries of L-glutamic acid (GA) in methanol. The results showed that SPCA can significantly improve the signal-to-noise ratio and hence lower the detection limits (approximately 0.389% g/mL for mannitol-methanol suspensions and 0.4 wt % for beta-form GA in GA-methanol slurries comprising mixtures of both alpha- and beta-forms of GA) thereby providing an important contribution to crystallization process performance monitoring.