NEURAL-NETWORK MODEL FOR PAPER-FORMING PROCESS

Paper is made by a continuous high-speed filtration drainage of an aqu eous suspension of fibers. This paper presents a new approach to the c ontrollable simulation of paper forming, using artificial neural netwo rk methods. The model incorporates dynamics of the forming process, li ke turbulence, drainage speed, and preferential drainage through earli er less-dense regions and fiber properties, like propensity to clump, or ''flocculate,'' fiber flexibility, and concentration of fibers in t he suspension. Results for monofiber layer structures are described, s howing effects of turbulence and its decay during drainage in causing clumping, or ''flocculation.'' The commercial process has, as one of i ts main goals, the reduction to tolerable levels of the nonuniformity in mass distribution resulting from flocculation. The new model yields data corresponding to that obtainable along arbitrary scanning lines in planar stochastic fibrous structures, providing profiles, variances , histograms of local areal density, and histograms of local free-fibe r lengths. These results closely resemble experimental data from comme rcial paper samples obtained from radiographic or optical transmission images subjected to image analysis.