As gigawatt-scale AI training campuses seek interconnection across the grid, utilities and interconnection engineers are running into a technical question that didn't feature prominently in traditional large-load studies. Historically, large-load studies centered on steady-state and transient stability concerns — thermal loading, voltage support, transient voltages and rotor angle stability — under the assumption that load demand varied slowly enough not to interact dynamically with nearby generation. AI training loads challenge that assumption.
Large loads with highly variable power demand profiles — the kind associated with AI model training — can introduce oscillatory behavior in load power consumption. That oscillatory signature is what makes this class of interconnection different from a traditional industrial load, and it's precisely what can trigger sub-synchronous oscillation (SSO) concerns on the power system.
When AI Load Meets Turbine Shafts
Turbine shafts on thermal units — steam and gas alike — are multi-mass mechanical systems with finite torsional stiffness between sections: the HP turbine, IP turbine, LP turbine, generator rotor, and exciter, among others. Rather than transmitting torque instantaneously as a rigid body would, each shaft segment behaves somewhat like a torsional spring between adjacent masses. That construction gives every unit a set of torsional natural frequencies, typically distributed between roughly 5 Hz and 50 Hz, and specific to each unit's particular mass and stiffness arrangement. Some units also exhibit super-synchronous natural frequencies, which generally damp out quickly without imposing additional stress on the shaft.
If an external oscillatory disturbance — such as a large load's variable power draw — lands at or near one of these natural frequencies, it can excite shaft torsional modes, much like a small periodic push at the right frequency can drive a mechanical system to swing at much larger amplitudes than the push itself. Under sustained or resonant conditions, this leads to cyclic mechanical stress, accelerated fatigue life consumption, and, in severe cases, shaft or coupling failure.
Why EMT Modeling and OEM Input Were Both Required
Quantifying this risk requires more than a standard phasor-domain interconnection study. Phasor-domain tools are well suited for voltage stability and generator swing analysis, but they aren't built to resolve oscillatory phenomena in the 5–50 Hz torsional range with the fidelity needed to assess shaft stress. An electromagnetic transient (EMT) platform is required to accurately model the large load, its profile, as well as the turbine shaft arrangement, since EMT simulations are what allow engineers to accurately quantify how power oscillations from the large load propagate to the generator turbine shaft.
Appropriately modeling the large load profile, the surrounding network, and the detailed turbine shaft multi-mass system in the EMT platform provides better quantification of the oscillations in the air gap torque caused by the large load's operation, as well as the propagation of those oscillations along the shaft — potentially exciting one or more of the shaft's torsional natural frequencies. Under such resonant conditions, the twisting torques, and hence the stress on the shaft, can become excessive.
A sample air gap torque measurement of a synchronous machine with oscillations caused by a large load is shown in Figure 1.
Figure 1: Air gap torque of the synchronous generator.
However, EMT results only get you to the torque and stress waveform — not the pass/fail answer. The stress level on the shaft for a given generator unit is material, geometry, and manufacturer specific, depending on forging material properties, stress concentration factors at couplings and shoulders, surface finish, and notch sensitivity at each shaft section. Two shafts subjected to an identical torque oscillation can see very different fatigue outcomes depending on these factors, which is why the equipment manufacturer of the affected thermal units should be engaged directly in the stress-level evaluation of the turbine shaft.
A sample plot of stress level along the shaft for a given oscillatory air gap torque is demonstrated in Figure 2.
Figure 2: Turbine shaft twist (top plot) and corresponding stress level along the shaft (bottom plot).
This pairing — system-level EMT simulation plus OEM-specific shaft mechanics — is what makes a study's conclusions credible and actionable rather than generic. As AI training campuses continue to seek interconnection near thermal and nuclear generation fleets, this combined approach is likely to become a standard part of the interconnection study toolkit rather than a specialized exception.
.jpg&w=3840&q=75)
.png&w=640&q=75)
