Currently, the integral windup bound is set to a fixed value. However, this can lead to overshoot e.g. during flow control when 100% power is needed to reach the target flow rate. This leads to e(t)*Kp + e_sum(t)*Ki = e(t)*Kp + 100 > 100 (where the last inequality holds only when the flow is below the target).
Instead of the fixed bounds, e_sum_lb <= e_sum(t) <= e_sum_ub, the windup bound on the integral term should dependent on the proportional input: e_sum_lb - e(t)Kp <= e_sum(t) <= e_sum_ub - e(t)*Kp. This is equivalent to the fixed bounds when at the set point, but it keeps prevents the error sum from growing when the error is large (i.e. the proportion term is dominate).
Currently, the integral windup bound is set to a fixed value. However, this can lead to overshoot e.g. during flow control when 100% power is needed to reach the target flow rate. This leads to
e(t)*Kp + e_sum(t)*Ki = e(t)*Kp + 100 > 100(where the last inequality holds only when the flow is below the target).Instead of the fixed bounds,
e_sum_lb <= e_sum(t) <= e_sum_ub, the windup bound on the integral term should dependent on the proportional input:e_sum_lb - e(t)Kp <= e_sum(t) <= e_sum_ub - e(t)*Kp. This is equivalent to the fixed bounds when at the set point, but it keeps prevents the error sum from growing when the error is large (i.e. the proportion term is dominate).