ABSOLUTE STABILITY AND DRIFT ROBUSTNESS OF A SECTOR-BOUNDED NEURO-FUZZY CONTROLLER FOR A DEAD-TIME COTTON SEED LINTING PROCESS

A saw linter that removes residual fibres from cotton seed is a dead-time plant whose gain depends on the operating point and falls as the saw set wears. This study checks whether a neuro-fuzzy controller for this plant stays stable over all operating conditions and the full saw-set life. The plant was modelled from industrial tests as a second-order lag with a dead time of 11.8 s; its gain changes 2.5-fold over the working speed range, and wear lowers it by 31.7% over a 248 h saw-set life. The controller is an incremental zero-order Takagi-Sugeno controller with 15 rules. Its input-output map was bounded by sectors, and stability was assessed by linear margins at the origin, a circle-type sector test on a 78 × 58 grid of operating points, a sensitivity study and simulation driven by an industrial record. Near the origin the phase margin is 78.6 to 86.2 deg and the gain margin 27.7 to 35.8 dB. The sector test is met on 99.7% of the domain and fails by at most 2.4% in a small corner of low saw speed and high feed rate, where simulation confirms stable operation. Wear lowers the loop gain and raises all margins: after about 18 h of operation the test holds on the whole domain. The binding limit is actuator saturation, which makes the 7.00% set-point unreachable after 142 h. The saw-set replacement time should therefore be set by the saturation share and the out-of-band share, not by stability.