On-line multivariate chemical data compression and validation using wavelets

Modern chemical plants produce a tremendous amount of measurement data that could be used to extract information for process monitoring and control. The data are typically transferred over the information highway and stored in a compressed form in a data historian. It is important to have an efficient technique for data compression such that the compressed data require minimum storage space and the reconstructed data retain the desired process features. Moreover, in an era where the process monitoring and control techniques are heavily dependent on the quality of data, it is imperative to have efficient data validation schemes. Also, the fact that it is undesirable to store spurious data makes sensor validation and compression important and related problems in process monitoring.; A wavelet based on-line data compression approach has been developed. Various analytical results on the bounds on compression ratio and sum of square error that can be achieved using this algorithm are derived. Experimental evaluation over two sets of plant data demonstrate the superiority of the proposed wavelet based approach over the conventional data compression methods.; A practical problem in implementing wavelet based methods for compression is that they need thresholding on the wavelet domain coefficients, which is not commonly intuitive to engineers. An error based criterion is proposed that uses semantically straightforward measures of the quality of the result to be obtained (such as the root mean square error (RMSE) and local point error (LPE)) to adaptively calculate the thresholds. Experiments show that the resulting algorithm gives superior compression as compared to other wavelet based methods.; Sensor validation is a pre-requisite for efficient process monitoring and control. Two sensor fault detection schemes based on (a) multi-scale analysis and dynamic Principal Component Analysis (PCA), and (b) multi-scale analysis of prediction model residuals, have been developed. The results obtained with these formulations outperform conventional PCA based evaluations, and yield low Type I (false alarm) and Type II (failure to detect faults) errors.; A multi-scale principal component analysis (MSPCA) technique is employed to detect and identify process faults and discriminate a normal plant operation from an abnormal situation. The proposed MSPCA approach is able to outperform the conventional PCA based approach in early detection and identification of actual process faults in an industrial data set.