Controle por aprendizagem iterativa aplicado a um modelo de ventilador mecânico para pequenos animais

Abstract

Mechanical ventilators are machines used to assist breathing and are widely used in research involving respiratory diseases. However, most commercial options available for small animals have limitations when tracking ventilatory profiles such as desired air pressure or flow. Iterative Learning Control (ILC) is a control technique that aims to improve performance of systems with repetitive tasks by learning from previous executions. This study proposes ILC control strategies for the problem of tracking profiles associated with ventilation modes. We use strategies based only in ILC, in a classical PI controller and in a combination of both. The control systems design was based on transfer functions, obtained from a simplified model of a feedback linearized recently proposed ventilator. The systems performance was evaluated with simulations, where we incorporated hypothetical scenarios with leakages, sensor noise and parametric uncertainty in the linearization. Considering that deviations from the reference profile can harm the patient’s lung, ILC alone was shown to be inadequate for mechanical ventilators in the initial iterations. The architecture with ILC and PI combined the benefits of both strategies, with an acceptable initial performance that still improves with each iteration. This strategy with PI and ILC remained robust for all the types of disturbances tested, especially in the scenarios with periodic leakage.