Dual variational model of a thermal breakdown of a dielectric layer at an alternating voltage (Q2001927)

The authors present a dual variational model for a thermal breakdown problem posed in a plane of circular cylindrical layer submitted to an alternative voltage. With the help of Kirchhoff's substitution and some change of variables and unknowns, they end with the elliptic problem \(\frac{1}{\rho^{n}}\frac{d}{d\zeta}(\rho^{n}\frac{d\Psi (\zeta)}{d\zeta})+\beta Z(\zeta)F(\Psi)=0\) for \(\zeta \in (0,1)\) with \(\rho =r/h\), \(\zeta =z/h\), \(h \) being the thinness of the layer, \(Z(\zeta)=1/(\rho \ln (1+1/\rho_{0}))^{2n}\), with \(\rho_{0}=r_{0}/h\), \(F(\Psi)=\tan \delta_{\ast}(\Psi) \). The parameter \(\beta \) is equal to \((\Delta U)^{2}f_{0}\varepsilon^{\prime}/(1.8\times 10^{10}h^{2})\) where \(\Delta U\) is the difference of the electrical potentials on both faces of the layer, \(f_{0}=\omega /2\pi \) is the AC voltage frequency and \(\varepsilon^{\prime}\) is the relative permittivity of the layer which is supposed to be constant. The boundary conditions \(\frac{d\Psi (\zeta)}{d\zeta}\mid_{\zeta =0}-F_{0}(\Psi_{0})=0 \) and \(\frac{d\Psi (\zeta)}{d\zeta}\mid_{\zeta =1}+F_{1}(\Psi_{1})=0\) are imposed where \(F_{0}(\Psi_{0})\) and \(F_{1}(\Psi_{1})\) are deduced from known temperatures functions \(F_{0}^{\ast}\) and \(F_{1}^{\ast}\) of \(T(0)\) and \(T(h)\), respectively. Multiplying the equation by a variation \(\delta \Psi (\zeta)\) and integrating by parts, the authors obtain the variational formulation \(\int_{0}^{1}(\frac{d\Psi (\zeta)}{d\zeta}\frac{d\delta \Psi (\zeta)}{d\zeta}-\beta Z(\zeta)F(\Psi)\delta \Psi (\zeta))\rho^{n}d\zeta +F_{0}(\Psi_{0})\delta \Psi_{0}+F_{1}(\Psi_{1})\delta \Psi_{1}=0\) and they observe that it is the stationarity condition of the functional \(J[\Psi ]=\int_{0}^{1}\left( \frac{1}{2}\left( \frac{d\Psi (\zeta)}\,{d\zeta}\right)^{2}-\beta Z(\zeta)\int_{0}^{\Psi (\zeta)}F(\Psi)d\Psi \right) \rho^{n}d\zeta +\rho_{0}^{n}\int_{0}^{\Psi_{0}}F_{0}(\Psi)d\Psi +(1+\rho_{1}^{n})\int_{0}^{\Psi_{1}}F_{1}(\Psi) d\Psi \). They introduce the condition \(Q(\zeta)+\frac{d\Psi (\zeta)}{d\zeta}=0\) for \(\zeta \in (0,1)\), the functional \(J[\Psi ,Q]=\int_{0}^{1}\left( \frac{1}{2}\left( Q(\zeta)\right)^{2}-\beta Z(\zeta)\int_{0}^{\Psi (\zeta)}F(\Psi)d\Psi \right) \rho^{n}d\zeta +\rho_{0}^{n}\int_{0}^{\Psi_{0}}F_{0}(\Psi)d\Psi +(1+\rho_{1}^{n})\int_{0}^{\Psi_{1}}F_{1}(\Psi)\,d\Psi \) and a Lagrange multiplier \(L(\zeta)\) for \(\zeta \in (0,1)\), which leads to the functional \( J[\Psi ,Q,L]=J[\Psi ,Q]-\int_{0}^{1}L\left( Q(\zeta)+\frac{d\Psi (\zeta)}{ d\zeta}\right) \rho^{n}d\zeta \). They compute its variation and they introduce an alternative functional \(I[\Psi ,Q]=J[\Psi ]-\frac{1}{2} \int_{0}^{1}\left( Q(\zeta)+\frac{d\Psi (\zeta)}{d\zeta}\right)^{2}\rho^{n}\,d\zeta \) and an integral error estimate \(J[\Psi ,Q]=\frac{1}{2} \int_{0}^{1}\left( Q(\zeta)+\frac{d\Psi (\zeta)}{d\zeta}\right)^{2}\rho^{n}\,d\zeta \). They observe that \(I[\Psi ,Q]\leq J[\Psi ]\) and that the equality occurs when the condition \(Q(\zeta)+\frac{d\Psi (\zeta)}{d\zeta} =0\) is satisfied. They then analyze the stationary points of the main functional \(J_{0}[\Psi ]=\int_{0}^{1}\left( \frac{1}{2}\left( \frac{d\Psi (\zeta)}{d\zeta}\right)^{2}-\beta \int_{0}^{\Psi (\zeta)}F(\Psi)d\Psi \right) d\zeta \).\ Taking \(F(\Psi)=A\exp (a\Psi)\), they derive from the above differential equation the expression \(\Psi_{0}(\zeta)=B_{0}-2\ln (\cosh (\sqrt{\beta \exp (B_{0})/2\zeta}))\). They plot the alternative functional \(I[\Psi ,Q]\) in terms of \(\Psi_{0}\) and they observe that the curve presents two extrema for which they analyze the stability behavior. The paper ends with the presentation of the simulations obtained on an example of such a situation.