A transformer core noise constraint links material strain to joints, supports, fluid and tank response. Build that chain into optimization to identify which candidate changes the governing acoustic mechanism. [1]
Write the acoustic requirement as a defined output under a defined duty. Sound power and sound pressure at a location answer different questions. Weighting, frequency content, cooling equipment and test environment must match the acceptance basis.
Follow the prediction chain
Begin with the magnetic field in the candidate core and magnetostrictive data appropriate to direction, induction and material state. Use those strains or the corresponding consistent forcing in a structural model. Evaluate the radiated acoustic response with the intended boundary conditions.
A published bare-core study coupled directional magnetic data, magnetostrictive strain, structural response and sound radiation, and compared the result with a core in air. That is useful evidence for the chain. Its validation does not extend automatically to an oil-filled tank, clamping changes or an assembled production transformer.
| Candidate screen | Condition for advancing | Reason to hold the candidate |
|---|---|---|
| Magnetic source model | Material and excitation data cover the duty | Strain behaviour inferred outside its evidence |
| Structural transfer | Supports, stiffness and damping are represented credibly | A changed clamp or joint inherits the old response |
| Acoustic output | Metric and environment match the stated comparison | Local pressure is compared with a power limit |
| Full-unit requirement | Validation covers the applicable construction and sources | Bare-core results are treated as unit acceptance |
Inspect which frequencies control the limit
Lower source strain does not guarantee a proportionate reduction in sound if a geometry change brings a structural mode closer to a forcing component. Retain the relevant spectrum and assess sensitivity to support and damping uncertainty.
Separate no-load core noise from load-related winding forces and auxiliary equipment noise when building the source model. Combine them using a method appropriate to the specified metric and source relationship; adding decibel numbers arithmetically is not a sound-energy calculation.
Compare the cost of passing the acoustic screen
A lower magnetic loading, different steel or changed support can affect mass, loss, excitation and temperature. Recalculate those outputs for the acoustic candidate. A noise remedy that obstructs cooling or creates an unacceptable magnetic condition is not a feasible optimum.
Use catalogue density 7,650 kg/m³ [2]. Assume participating steel volume 0.010 m³, modal mass 76.5 kg, stiffness 30 MN/m and damping ratio ζ = 0.03. Natural frequency is 99.67 Hz; adding 10% modal mass at unchanged stiffness gives 95.03 Hz. At an assumed 50 Hz electrical duty, take a dominant 100 Hz forcing component. For this independent single-mode screen, hold force, damping and acoustic radiation transfer unchanged. Dynamic amplification D = 1/√[(1−r²)²+(2ζr)²], r = 100/fn, falls from 16.51 to 8.03. With sound power proportional to squared surface velocity at that same frequency, the change is 20 log₁₀(8.03/16.51) = −6.26 dB. Assume this mode dominates total unweighted sound power, initially 70 dB re 1 pW, with a 68 dB project limit. The mass variant predicts 63.74 dB and passes that assumed screen; the baseline fails. Advance it for coupled validation, verifying the unchanged-force and radiation assumptions before claiming a transformer noise margin. [1]
Clamp stiffness, preload, joint registration and support contacts control excitation and vibration transfer. Record those parameters with the noise result; changing steel grade without stabilising them may leave the acoustic limit unresolved.
References
- R. Haettel, M. Kavasoglu, A. Daneryd and C. Ploetner, ABB, Prediction of Transformer Core Noise, COMSOL Conference, Cambridge, 2014.
- Nippon Steel Corporation, Grain-Oriented Electrical Steel Sheets, catalogue D004je, April 2020, specification and typical-property tables.