This can sometimes happen if the region bounds intersect with the region it self. If you look at the output of the ElementMesh you will see that it's bounds do not go from -radius to radius.
mesh["Bounds"]
{{-0.995306, 0.995306}, {-0.995306, 0.995306}, {0., 3.}}
The fix is easy: just enlarge the bounding box a bit:
height = 3;
radius = 1;
thickness = .06;
rt = radius + 0.1;
Needs["NDSolve`FEM`"]
ir = ImplicitRegion[(radius - thickness)^2 <= x^2 + y^2 <= radius &&
0 <= z <= height, {x, y, z}];
mesh = ToElementMesh[ir, {{-rt, rt}, {-rt, rt}, {0, height}},
"MeshOrder" -> 1, MaxCellMeasure -> 0.0003];
Look at the bounds:
mesh["Bounds"]
{{-1., 1.}, {-1., 1.}, {0., 3.}}
Look at the mesh:
mesh["Wireframe"]

Update:
Another option is to manually generate the a hex element based mesh for this case:
Needs["NDSolve`FEM`"]
nx = 100; ny = 5; nz = 100;
coordinates =
Flatten[ Table[{r Cos[\[Theta]], r Sin[\[Theta]], h}, {h, 0.,
3., (3 - 0)/(nz - 1)}, {r, 1. - 0.05,
1., (1. - (1 - 0.05))/(ny - 1)}, {\[Theta], 0.,
2 Pi, (2 Pi - 0.)/(nx - 1)}], 2];
mkIncidents =
Compile[{{nx, _Integer, 0}, {ny, _Integer, 0}, {nz, _Integer, 0}},
Flatten[
Table[Block[{p1 = (j - 1)*nx + i, p2 = j*nx + i, p3 = p2 + 1,
p4 = p1 + 1, p5, p6, p7, p8},
{p5, p6, p7, p8} = {p1, p2, p3, p4} + k*nx*ny;
{p1, p2, p3, p4} += (k - 1)*nx*ny;
{p1, p2, p3, p4, p5, p6, p7, p8}], {i, 1, nx - 1}, {j, 1,
ny - 1}, {k, 1, nz - 1}], 2]
];
incidents = mkIncidents[nx, ny, nz];
mesh =
ToElementMesh["Coordinates" -> coordinates,
"MeshElements" -> {HexahedronElement[incidents]}]
ElementMesh[{{-0.999497, 1.}, {-0.999874, 0.999874}, {0.,
3.}}, {HexahedronElement["<" 39204 ">"]}]
mesh["Wireframe"]

Playing with the nx,ny and nz allows you to influence the the quality which you can inspect with:
Histogram[mesh["Quality"]]
thickness = 0.01? I either get an error, or I can reduceMaxCellMeasurebut it never ends. – anderstood Feb 02 '18 at 18:03MaxCellMeasure -> 0.0000015it gives me 1.3M tets. I am a bit surprised that I need to specify the MaxCellMeasure, but I assume because of the high aspect ratio the automatic MaxCellMeasure computation fails. This does not look quite correct. – user21 Feb 02 '18 at 19:16