(202u) Application of the Wavelet Transform for Process Data Compression

Improvements in sampling methods and sensor technology have resulted in the creation of vast amounts of process data that must be compressed and stored. Data compression using a wavelet transform results in minimum information loss and has been commonly used for image compression and batch data applications. In this study, a discrete wavelet transform was used to compress and then reconstruct several different types of process data (temperature, flow rate, valve control). Data sets were converted into sets of approximation coefficients and detail coefficients using a predetermined threshold and down-sampled repeatedly up to a chosen decomposition level. The signal was then reconstructed and compared with the original signal using root mean square error (RMSE) and local point error (LPE). Both values were found to significantly decrease with increasing decomposition level. The exponential decrease in error also allowed for the determination of an ideal decomposition level and compression ratio for each signal. The reconstructed data was found to effectively follow the trends of the original data while also having a denoising effect. The wavelet transform method was also successful in eliminating artificial "faults" (zero values) that were introduced into the data. However, data had to be truncated to consider boundary effects caused by wavelet transforms. The wavelet transform algorithm was therefore found to effectively compress and reconstruct the majority of the signal. These results will be compared with results from the PI software by OSI Soft to determine the feasibility of the wavelet method. The PI software applies the 'swinging door algorithm' and is commonly used in industry.