AOO-ANFIS: A Novel Bio-Inspired Neuro-Fuzzy Controller for Hydrostatic Transmission Stroke Control

Hydrostatic transmissions (HSTs) rely on coordinated pump and motor displacement control to achieve smooth stroke regulation. This study proposes AOO-ANFIS, a bio-inspired neuro-fuzzy controller in which the 66 premise and consequent parameters of a two-input, two-output first-order Takagi–Sugeno adaptive neuro-fuzzy inference system (ANFIS) are tuned by the animated oat optimization (AOO) algorithm rather than by ANFIS’s conventional gradient-based hybrid learning rule. The controller is trained as a data-driven surrogate for a previously validated multiple model predictive controller (MMPC): closed-loop input–output samples generated by the MMPC are used as supervised targets, enabling the resulting fuzzy controller to reproduce the MMPC control policy without solving an online optimization problem. On this identical tuning task, AOO was benchmarked against Harris Hawks Optimization (HHO), particle swarm optimization (PSO), a real-coded genetic algorithm (GA) and the covariance matrix adaptation evolution strategy (CMA-ES). It reached the lowest mean validation error, significantly lower than HHO, GA and CMA-ES and statistically indistinguishable from PSO. The AOO-ANFIS controller was subsequently deployed in a processor-in-the-loop simulation on a BeagleBone Black target, replacing the previously reported MMPC under identical deterministic and stochastic torque disturbance profiles. Relative to the MMPC, the AOO-ANFIS controller reduced the maximum overshoot from 25% to 14% under deterministic disturbances and from 24% to 14% under stochastic disturbances, reduced the closed-loop tracking RMSE by 7.8% and 30.1%, respectively, and reduced the integral squared-error indices ISE and ITSE by 15% and 21% under deterministic and by 51% and 41% under stochastic disturbances. It did so with an average control effort about 10 percentage lower and an average computational load 76–84% lower, at the cost of a longer peak time under stochastic disturbances: 1.8 s instead of 1 s. In an extended 300 s scenario containing reference levels, ramp rates, a sinusoidal segment and trapezoidal disturbances absent from the training data, the AOO-ANFIS maintained bounded tracking with an RMSE of 0.035 over the unseen segment, lower than that of a previously published ANN surrogate trained on the same data, 0.056. Both surrogates remained stable when the plant’s volumetric-efficiency maps were degraded by about 15%, with the AOO-ANFIS showing lower error. Moreover, this is the first application of the AOO algorithm to ANFIS tuning and the first bio-inspired-optimization-tuned ANFIS controller for HST stroke control.