Time Optimization of Constrained Control for a Thermoelectric Solid System with a Peltier Element

TL;DR

This paper develops a numerical method for time-optimal constrained control of a thermoelectric system with a Peltier element, achieving faster steady-state temperature distribution through feedback linearization and eigenfunction decomposition.

Contribution

It introduces a finite-dimensional approximation and a gradient descent approach to compute optimal piecewise constant control laws for the system.

Findings

Control law significantly reduces transient time to reach steady state.

Eigenfunction decomposition enables efficient numerical solution of the control problem.

The method effectively handles electric power constraints in the control design.

Abstract

A solid system consisting of two heat conducting cylinders with a thermoelectric converter (Peltier element) between them is considered. A nonlinear model, which was previously verified by authors, is used to design a constrained control law that allows us to achieve a steady-state distribution of the temperature in one of the cylinders in much less time than the characteristic time of transient processes. The initial-boundary value problem is exactly linearized over temperature by means of feedback linearization. Although the resulting system is nonlinear in a control function, it is possible to construct a finite-dimensional approximation based on analytical solution of the corresponding eigenproblem for a constant control signal. The time-optimal control problem is studied numerically by using this eigenfunction decomposition. To construct admissible control laws, an auxiliary unconstrained optimization problem is introduced. Its cost functional represents a weighted sum of temperature deviation from the desired zero distribution and a penalty for violating an electric power constraint. The control time interval is split into several parts, and on each subinterval the control signal is taken constant. The optimal piecewise constant feedforward control is found numerically by applying the gradient descent method. We analyze the proposed control law with respect to the shortest admissible time of the process.