Rapid Detection Of Physical And Chemical Properties Of Fermentation Broth Based On Hyperspectral Imaging Technique

The amount of and variation of the feedstock has large impact on the stability of the anaerobic digestion process, detecting timely the amount and variation of the solid feedstock in the digester plays a significant role in the control of the whole process. Traditional laboratorial analysis based on chemical principles is time-consuming and needs professional personnel which makes it difficult to satisfy the practical demand. Spectral analysis is a very promising process analysis technique (PAT) to achieve the purpose of in situ monitoring due to its fast and simple attribute. In this paper, hyperspectral coupled with chemometrics and imaging analysis methods are applied to establish the calibration and validation models for total solid, volatile solid, pH and total inorganic carbon concentration so as to lay the foundation for deploying the hyperspectral system in practice.(1) which are partial least square (PLS), least square-support vector machine (LS-SVM) and three wavelength selection methods, which include adaptive weighted sampling competition (CARS), random frog (RF) and Successive Projections Algorithm(SPA), are applied to predict the solid concentration of anaerobic digestion solution. The root mean square error prediction and correlation efficient of the total solid concentration are respectively 0.0058g/L and 0.841; the root mean square error prediction and correlation efficient of the volatile solid concentration are respectively 0.0041g/L and 0.874.(2) The calibration and validation methods for pH and total inorganic carbon concentration (TIC) are established. Firstly, a hyperspectral imaging system covering the spectral range 874-1734 nm was used to determine pH value of anaerobic digestion liquid produced by water hyacinth and rice straw mixtures for methane production. Wavelet transform (WT) was used to reduce noises of the spectral data. Successive projections algorithm (SPA), random frog (RF) and variable importance in projection (VIP) were used to select 8,15 and 20 optimal wavelengths for pH value(3) prediction, respectively; Competitive adaptive reweighted sampling(CARS), SPA and RF were used to select 14,11,8 optimal wavelengths for TIC concentration spectral data. Partial least squares (PLS) and back propagation neural network (BPNN) were used to build calibration models on full spectra and the optimal wavelengths for pH and TIC. As a result, BPNN models performed better than the corresponding PLS models, and SPA-BPNN model obtained the best performance with correlation coefficient of prediction (Rp) of 0.911 and root mean square error of prediction (RMSEP) of 0.0516 for pH; and even though full spectral PLS outperformed RF-BPNN slightly, the models based on optimal wavelengths are more robust and efficient, therefore RF-BPNN models were used and the Rp and RMSEP were 0.733 and 133.359 mg/L.(4) The models SPA-BPNN of pH and RF-BPNN of TIC built earlier are applied to predict the pH and TIC concentration of each pixel within the hyperspectral images and the distribution maps of pH and TIC were achieved. The steps are (i) the hyperspectral images of anaerobic digestion are masked by using the ENVI, making sure the reflectance ratio is 0 by getting rid of the background images (ii) the images are smoothed and denoised by applying the WT algorithm (iii) the distribution maps of pH and TIC are achieved by applying the models in the Matlab. The result of pH distribution map is quite satisfactory while the TIC map is not as well as expected. All in all, the visualization maps provide intuitive experience for the user to detect the parameters of the anaerobic digestion liquid.The study indicates that utilizing hyperspectral combined with chemometrics methods to predict the solid concentration of the anaerobic digestion solution is feasible, it can provide a theoretic and practical basis for detecting the solid concentration of anaerobic digestion process.