Seems like it would work - https://www.youtube.com/watch?v=xEK-c1pkTUI

The dye is non reactive and operates independently of being a metallic part.

A method that won't work is magenetic particle inspection as it depends on the ability of metal to concentrate the magnetic field at discontinuities and ceramics are usually non-magnetic.

Wouldn't the dye tend to form arc tracks if it can't be completely cleaned off?

You would need to use a dye with very high dielectric properties.

= = = = = = = = = = = = = = = = = = = =
P.E. Metallurgy, consulting work welcomed

Ultrasonic methods would probably work, as long as the crack was not at some specific hard-to-reach angles within the structure.

Converting energy to motion for more than half a century

A Doble (or similar) Power Factor / Tan Delta "Hot Collar" test has been used for decades to find cracked or broken insulators. If you don't have a test set, there are plenty of service companies that will.

From a NETA article on testing:

"Hot Collar Test
A hot collar test generates a localized high-voltage stress by using and energizing a conductive collar around the porcelain. This can determine dielectric losses (watts and current, as PF is not analyzed) through various sections of any bushing when troubleshooting high PF results. It is useful to detect cracked porcelain, deteriorated cement joints, gasket leaks, or faults within condenser layers in condenser-type bushings, as well as to check the oil level of oil-filled bushings when normal oil level readings have been established or when comparing to bushings of the same type. If an abnormal mA or watts reading is obtained when the test is performed under the top petticoat, the test should be repeated under the second petticoat and move further down until normal readings are found.

The hot collar test can be performed with any of the test modes — GST, GST-g, UST — using one or multiple collars depending on the anticipated outcome of the troubleshooting (Figure 7 from Paul Gill’s Electrical Power Equipment Maintenance and Testing).

Guidelines for evaluating hot collar data:

• Watts loss

– <100 mW: Acceptable

– ≥100 mW: Unacceptable, possible contamination

• Current

– Within 10% of similar bushings: Acceptable

– <10% of similar bushings: Unacceptable, low level of liquid or compound, possible faults within condenser layers
"