Numerical Field Computation

By the end of Chapter 2 we could write the stress at any point inside a coaxial cable or a pair of concentric spheres — exactly, in a single line of algebra. Those wins were real, but they rested on a luxury that real apparatus rarely grants: perfect symmetry. The instant a geometry departs from the clean cylinder or sphere — a bushing with grading rings, a post insulator with sheds, a circuit-breaker contact, a void buried in cable insulation — the tidy integrals collapse and no formula survives.

This is not a failure of the physics. The governing law is exactly the same as before; the field still satisfies Laplace's equation in the insulation and still answers to the electrode voltages on the boundary. What fails is our ability to integrate it by hand over an awkward shape. So we change tactics. Instead of seeking one continuous formula, we chop the problem into a great many small, simple pieces, write down the easy local rule each piece must obey, and let a computer juggle the thousands of resulting equations until they all agree. The continuous unknown becomes a long list of numbers, and solving the field becomes solving a large system of linear equations.

Four such schemes dominate high-voltage practice, and they split neatly into two philosophies. Finite difference and finite element methods fill the whole insulating volume with a mesh and solve for the potential at every mesh point. Charge simulation and boundary element methods leave the volume empty and work only on the electrode surfaces. Each philosophy has its natural home, and a good engineer knows which to reach for — and how to check that it told the truth.

Every method in this chapter is, underneath, an attempt to solve one boundary-value problem. In a region free of space charge the potential obeys Laplace's equation; where free charge \(\rho\) is present — an ionised region, an injected space charge — it obeys the more general Poisson's equation:

A differential equation alone has infinitely many solutions; what pins down the field is the set of conditions on the boundary of the region. Three kinds matter in high-voltage work. A Dirichlet condition fixes the potential — every conductor is an equipotential at its terminal voltage. A Neumann condition fixes the normal derivative — along a plane of symmetry, or a far-field boundary, the field has no component crossing it, so \(\partial V/\partial n = 0\). And at the interface between two different dielectrics, two matching conditions hold simultaneously:

The second condition is the reason field lines kink as they cross from oil into a solid spacer, and it is exactly what the capacitance grading of Chapter 2 exploited. Once a method has found \(V\) everywhere subject to all of these, the stress comes from the same gradient as always, \(\vec{E} = -\nabla V\), and the engineer reads off \(E_{\max}\) at the worst point. Everything that follows is just four different ways of finding that \(V\).

The finite difference method is the most intuitive of the four, and the best place to build a feel for what "discretising a field" really means. Lay a regular square grid of spacing \(h\) over the region. At each grid point we will store one number — the potential there — and our job is to find all of them at once. The trick is to rewrite the derivatives in Laplace's equation as differences between neighbouring grid values. A Taylor expansion of \(V\) in both directions gives the standard second-difference approximations, and adding them turns \(\nabla^{2}V = 0\) into a purely algebraic statement linking each node to its four neighbours:

That result is worth pausing on. It says the potential at any free node is simply the average of its four neighbours — a beautifully physical statement of "no charge here, so nothing piles up." When space charge is present the rule gains one term, \(V_{0} = \tfrac14(V_E+V_W+V_N+V_S) + \tfrac{h^{2}\rho}{4\varepsilon}\), the discrete form of Poisson. We fix the boundary nodes at their known electrode potentials and then sweep through the interior again and again, replacing each node by the current average of its neighbours; the grid relaxes toward the true solution. Convergence is hastened by over-relaxing — nudging each node a little past the plain average:

The finite element method keeps the idea of filling the region with a mesh but throws away the rigid grid. Instead it tiles the domain with small triangles (in 2-D) or tetrahedra (in 3-D) that can be made tiny and dense where the field is fierce and coarse where it is calm. Because the triangles can take any shape, the mesh hugs a curved electrode exactly — no stair-stepping. Within each element the potential is assumed to vary in a simple known way, usually linearly, described by shape functions \(N_i\) that interpolate between the element's corner values:

To choose the corner values, FEM invokes a global principle rather than a local stencil: the true field is the one that minimises the electrostatic energy stored in the region (equivalently, it makes the weighted residual of Laplace's equation vanish). Writing that energy as a sum over all elements and demanding it be stationary produces one equation per node, assembled into a large but very sparse matrix:

Solving that system delivers the potential at every node, and dielectric interfaces are handled almost for free because each element simply carries its own \(\varepsilon\). This blend of geometric flexibility and clean treatment of mixed materials is why FEM is the engine inside almost every commercial field solver, and why it is the default tool for analysing a modern graded bushing or a GIS spacer.

The charge simulation method takes a completely different view, and it is the one high-voltage engineers reach for most often when electrodes sit in open air. Its insight is this: the real surface charge spread over an electrode produces some field; if we could replace that smeared charge with a modest number of simple fictitious point or line charges, tucked just inside the conductor, we could reproduce the same external field by superposition — and then evaluate it anywhere with nothing more than Coulomb's law.

The recipe is direct. Place \(n\) fictitious charges inside each electrode, and choose \(n\) contour points on its surface. Demand that the charges, summed up, reproduce the known electrode potential at every contour point. Each such demand is one linear equation, with the potential coefficients \(P_{ij}\) (the potential at point \(i\) due to a unit charge \(j\)) as the matrix:

Solve that small system once for the charge magnitudes and the field at any point \(P\) — including right at the most-stressed spot — follows from a direct sum, no differentiation of a meshed potential required:

Two properties make CSM a favourite in high-voltage analysis. First, because empty space is never meshed, an electrode in open air — a transmission tower, an outdoor terminal — is handled as naturally as one in a box; the field simply decays to zero at infinity on its own. Second, the field comes out analytically and so is very accurate near sharp electrodes, precisely where the volume methods strain. The cost is bookkeeping: a separate set of charges is needed for each dielectric region, and placing the charges well is something of an art.

The boundary element method shares CSM's surface-only spirit but rests on firmer mathematical ground. Rather than guessing where to drop fictitious charges, it recasts Laplace's equation as an integral equation over the boundary using Green's functions, then discretises only that boundary into small elements. The unknowns live solely on the electrode and dielectric surfaces, so a three-dimensional problem collapses to a two-dimensional mesh — a large saving — while open regions are again handled gracefully. The trade-off is a dense, fully populated matrix, in contrast to the sparse one FEM produces.

Beyond these four, a few specialised tools round out the field. The finite volume method, dominant in fluid dynamics, enforces conservation cell by cell and appears in coupled electro-thermal studies. Monte Carlo techniques estimate the potential at a single point by random walks, occasionally useful for spot checks. For the overwhelming majority of high-voltage work, though, the choice is between the volume pair (FDM, FEM) and the boundary pair (CSM, BEM) — which brings us to how an engineer actually decides.

There is no single best method, only a best fit for the problem in hand. The decision usually turns on three questions: is the region bounded or open, how many different dielectrics are involved, and how sharply does the field peak? The table below distils the working trade-offs.

Whatever the choice, the discipline that separates a result from a guess is validation. Before trusting a solver on a bushing whose answer you cannot check, run it on a geometry whose answer you can — the coaxial cable or the concentric sphere of Chapter 2 — and confirm it reproduces the exact \(E_{\max}\) to within a fraction of a percent. Then refine: halve the mesh spacing, or add more charges, and watch the answer settle. When successive refinements stop changing the result, it has converged.

For each item, first name the method or condition it tests — the governing equation, a boundary condition, the FDM stencil, the FEM idea, the CSM system, or validation — then work it through. Difficulty rises down the list.