Impact of Data Compression and Quantization on Data-Driven Process Analyses
- PDF / 1,260,609 Bytes
- 21 Pages / 441 x 666 pts Page_size
- 58 Downloads / 212 Views
Impact of Data Compression and Quantization on Data-Driven Process Analyses
Modern chemical plants are extensively instrumented and automated. One manifestation of increasing automation and accessibility is easy availability of process information on the desktop. Information access is typically available as a result of extensive data logging and process archiving. Historical data are an invaluable source of information, but in chemical industrial practice, data are often compressed using various techniques, e.g. box car, backward slope, swinging door, PLOT, wavelet transform, etc. before storing them in a historian. Compression degrades data quality and induces nonlinearity. This chapter focusses on the problems of data quality degradation and nonlinearity induction due to compression of the process data. It also presents an automatic method for detecting and quantifying the degree of compression present in the archived data. Finally, the problem of quantization in the process data is discussed, and an automatic procedure to detect and quantify quantization is presented.
4.1 Introduction Motivations for data compression discussed in Thornhill et al. (2004) included reduction of the costs of storage of historical data and reduction of cost of transmission of process data through a telecommunications link. The same paper discussed the hidden costs of data compression, in particular that the data often become unsuitable for their intended purposes. Desired end uses of the data surveyed by Kennedy (1993) include: • • • • •
Calculation of daily statistics such as daily means and daily standard deviations; Averaging for data reconciliation and mass balancing; Archiving of data trends for subsequent high fidelity reconstruction; Data smoothing by removal of high-frequency noise; Feature extraction and recovery of events.
Once the data have been compressed, however, they lose information and the reconstructed trends are deficient in various ways compared to the originals. Figure 4.1 shows an example of a data set that has been reconstructed after compression in a
M. A. A. S. Choudhury et al., Diagnosis of Process Nonlinearities and Valve Stiction, c Springer-Verlag Berlin Heidelberg 2008
45
46
4 Impact of Data Compression and Quantization on Data Driven Process Analyses
Tag 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 0
250
500
750 1000 time/sample interval
1250
1500
Fig. 4.1 An industrial data set with compression in some tags. Time trends are mean-centred and normalized
data historian. The straight line segments that are characteristic of industrial data compression can be seen in many of the time trends. The original uncompressed data were lost forever when they were compressed and archived, and it is now not possible to determine what features have been lost. This chapter first describes methods in common use for data compression and shows some examples of the adverse impact of data compression on data-driven plant performance analysis. It also outlines the compression detection algorithm th
Data Loading...
