Inferential relationships in thermomechanical pulp refining, explored using multivariate analysis.

摘要

The pulp and paper sector, like many globalized industries, finds itself with an increasingly demanding clientele who continually expect a better and cheaper product. An important design strategy for addressing this objective is the analysis of the vast quantity of process and product data accumulated in plant-wide data historians.; Using a Thermo-Mechanical Pulp (TMP) newsprint mill in Eastern Canada as a case study, we compared various ways of selecting and pre-treating raw process data to maximize the realism and usefulness of the black-box pulp quality models. A major conclusion was that models generated using an MVA variant known as Partial Least Squares (PLS) were significantly improved by pre-treating the data, with respect to both statistical significance and physical interpretability. We recommended an overall approach for applying MVA to industrial operating data, involving a systematic method for removing dubious periods of operation such as low production and aberrant process behavior, and filtering of all variables.; The next challenge was to develop a method for correlating TMP operations with pulp and, ultimately, paper quality by focusing on process fundamentals such as specific energy and refining intensity. The case study mill experiences variability in paper strength and porosity, which are known to be affected by upstream pulp refining conditions. Frequent interruptions in the refining lines greatly affect the specific energy and other key parameters, some of which are not measured directly and must be calculated from other variables. Due to the infrequency of measurements, linking paper variations to upstream pulp quality variables such as freeness, fiber length and fines content is difficult. However, the mainline refining and reject sections are 'data rich', with many parameters measured second-by-second. Unfortunately, data on the incoming wood chips were extremely sparse, and were found to contribute little to the MVA models.; A very important, and sometimes overlooked, part of any TMP mill is reject refining. Because the case study mill experiences frequent starts and stops on the four main refining lines, the throughput at the presses, and ultimately the reject refiners, is continually changing. Partly to compensate for this, the operators increase the reject rate after a line stoppage, which changes the fibre length distribution of the pulp entering the reject refiners. Combined with occasional stoppages of the reject refiners themselves, this situation results in reject refining that is highly variable and less well controlled.; Using Multivariate Analysis and other statistical tools, it was possible to link product quality back to TMP and rejects operations, taking into account number of lines in operation, plate age and process lags. The best MVA models were obtained by using a shorter (1-hour) data timescale, with a weighted-average filter helped to bridge the gap between the faster readings and slower paper quality trends.; Based on the above results, we have developed an explicit, detailed methodology for applying MVA to production data which addresses these challenges. In summary, the methodology involves defining the modelling objectives, examining product data, building a suitable model structure using process fundamentals, pre-treating the data sets, creating models with MVA, interpreting the statistical results, and finally identifying the limitations of the models.; Using this methodology, it was possible to use MVA to correlate roughly half of the variability in final paper quality with the refining operations. The upstream variables that were the most prominent in the models were: plate age; specific energy, including reject refining energy, which had to be calculated from other variables; refining consistency, which also had to be calculated; number of TMP lines in operation; and variability in motor load, as represented by the standard deviation. The case study thus yielded results t