Abnormity Identification In Process Of Pulpwood Properties Detection By Near Infrared Spectroscopy

Rapid detection of physicochemical properties of pulpwood in pulping process can be used to adjust digesting process parameters in real time,ensure the quality of pulp products,and improve the intelligent manufacturing level of pulping and papermaking industry.As a rapidly developing fast detection technology,near infrared spectroscopy(NIRS)has obvious advantages in detecting the components in natural products,and can be applied suitably to determine the pulpwood properties.To aim at the promotion of the NIRS technology in pulpwood properties detection,several common pulp woods and their NIRS were investigated dedicatedly.Some abnormity identification algorithms were designed to ensure the validity of information sources in the detection process of pulpwood properties.The main research work and conclusions are summarized as follows:(1)Discrimination of abnormal spectra.The abnormal spectra of glitch,cliffs and white noise caused by disturbance can be identified by high frequency signals decomposed from the original signal by wavelet transform.The white noise abnormal spectrums were simulated by adding different intensity white noise and the abnormal spectrum whose SNR is higher than 100 dB after denoising is repaired by means of wavelet transform filtering.The algorithms identifying outlier spectrum with severe drift during the parallel measurement is proposed for abnormity elimination to ensure the accuracy of pulpwood sample spectrum.(2)Discrimination of outlier samples in calibration set.Three abnormal sample detection methods based on Mahalanobis Distance(MD),Cross Validation(CV)and Monte Carlo Cross Validation(MCCV)are investigated to identify and eliminate abnormal samples which affect the performance of the model.The validation results show that the detection strategy based on MD and MCCV are suitable for discrimination of abnormal samples in calibration set of pulpwood,and the detection methods based on CV will lead to misjudgment,which leads to the performance degradation of the model.(3)Discrimination of outlier samples in predicting analysis process.Based on the information of the calibration set and the characteristics of the pulpwood samples for inspection,the type of outlier sample is determined by the criterion of MD as the out of bounds samples of concentration confined by calibration set.The comparison of two dimension reduction methods for MD calculation,PCA and PLS,was also given.The experimental results show that the outlier sample detection method based on MD can effectively identify out of bounds samples which would not be predicted correctly by the established pulpwood analysis model of holocellulose and lignin,and the dimension reduction method of PCA is more accurate than the PLS for the identification of out of bounds samples.