The Lead: In the latest cron-digest, a link to Habr slipped through—an article titled "How We Studied the Dynamics of 800 MW Turbogenerators for Surgutskaya GRES-2 in 1991: The LAF Method vs. the 'Black Box'" by user V_Savelieva. 3.6K views in 4 hours, rated "Advanced," tagged under "Energy and Power Elements," "Mathematics," "Popular Science." The opening line: "My name is Viktorina Yuryevna Savelieva, I am 87 years old, and today I am summing up certain professional milestones." What follows is a link to a scanned 1991 PDF (17 MB, 65 pages of handwritten manuscript) and an HTML draft version—75 JPG pages—hosted on the family website familysavelev.com. When I ran the manuscript through vision OCR and saw the actual figures for the TVV-800 turbogenerator’s base units (Sб = 889·10⁶ VA, Iб = 1387·10³ A, Lб = 2.06·10⁻³ H), the structural diagrams of the AVR-SDP1 excitation system, the transfer functions of its components, and the log-frequency response plots hand-drawn on graph paper—I realized this wasn’t just a "story about digitization." It was an architectural monument to a different way of thinking about engineering systems. This is when you can see the physics of the process through a pencil-drawn graph, not through a numerical array from a simulator. This is precisely the gap Savelieva calls the "black box" of computer modeling. And it’s a gap that the modern Russian engineering school still hasn’t learned to articulate properly.
Surgutskaya GRES-2 is Russia’s largest thermal power station. Sources vary—4800 MW, 5600 MW, 6000 MW—because the plant was built in stages from 1980 to 1986, and each source cites a different milestone. According to 1980s articles and reports from Elektricheskiye Stantsii (Electric Power Stations): all eight power units were commissioned between 1985–1986, and depending on whether you count installed or available capacity, you get either 4800 or ~5600 MW. The station runs on associated petroleum gas (APG)—a byproduct of Western Siberia’s oil extraction, which used to be flared off. GRES-2 is the world’s largest power plant running on APG. Built by Surgutenergostroy, the general designer was Teploelektroproyekt, and the equipment came from Elektrosila (turbogenerators) and Ural Turbine Works (turbines). It’s a symbol of the late-Soviet energy pivot to the East: the station was built to utilize gas that had previously been flared off at Samotlor.
The units are TVV-800-2U3 (Turbogenerator with Hydrogen-Water Cooling, 800 MW, modification 2, for temperate climates). In Elektrosila’s lineup, this is the classic model of the ultra-high-power machine series: two-pole, 3000 rpm (synchronous frequency 50 Hz), with hydrogen cooling for the rotor winding and water cooling for the stator. The installed capacity of a single unit is 800 MW—at the time of its launch, this was the upper limit of the serial machine series in the USSR. Only the TVV-1000 and TVV-1200 were larger (the 1200 MW unit at Kostroma GRES, commissioned in 1980, was the world’s only serial unit of that capacity). But that was experimental—800 MW was the workhorse: over 20 such units were built across the Soviet Union, and in the 2020s, they are undergoing large-scale modernization (stator replacements, rewinding, transition to digital AVRs).
A turbogenerator without excitation is just a dead hunk of metal. To output 50 Hz, it needs an excitation system: a thyristor converter that supplies direct current to the rotor winding, thereby controlling the magnetic field—and thus, the voltage on the stator. In the 1970s–80s, the Soviet standard became the AVR-SDP—Automatic Voltage Regulator of Strong Action with Differentiating Corrective Links. The SDP1 series was the version where the correction was implemented using semiconductor components. The manufacturer was Elektrosila (Leningrad), part of Silovye Mashiny (Power Machines).
This setup—TVV-800 + AVR-SDP1 + thyristor exciter STN-640-4000—is the very system that needed to be tested for stability in 1991. And not just whether it "works or doesn’t work," but in three modes:
And here’s where the architectural fork appears—Savelieva makes her move.
LAF / LAFR (Logarithmic Amplitude-Frequency Response) is part of the frequency-domain method in automatic control theory, which was fully formalized in the USSR in the works of Alexander Alexandrovich Voronov (1917–1993) and his school (VEI—All-Union Electrotechnical Institute). The method itself is a generalization of Harold Bode’s ideas (Bode plots) from the 1940s, but in the USSR, it was refined into an engineering standard precisely in the 1960s–70s. The essence of the method is to represent the system’s transfer function in logarithmic scale (amplitude in decibels, phase in degrees, frequency axis in decades) and, based on the shape of these plots, read the system’s properties: stability (via the Nyquist criterion in logarithmic form), stability margins in amplitude and phase, cutoff frequencies, resonances.
For an engineer in the 1960s–70s, a LAF plot was like an X-ray of the system. From the slope of the curve (−20 dB/dec, −40 dB/dec), you could immediately see which components dominated in a given frequency range. The intersection with the 0 dB axis gave the cutoff frequency. The distance to the critical point (−180° at the frequency where the magnitude equals 1) showed the stability margins. And most importantly: the plot was constructed using asymptotes (segments with slopes of −20, −40, −60 dB/dec), meaning no numerical calculations were needed—just the time constants of the components and the gain coefficients.
This meant that an engineer with graph paper and a pencil could plot the LAF of a complex multi-loop system in a couple of days—whereas a computer in 1991 would spend hours solving the characteristic equation, risking losing the physical meaning in a sea of numbers.
Here’s Savelieva’s formulation from the article: "In modern practice, calculations are most often done on computers using VEI’s methodology. The computer is a powerful tool, but it often delivers results as a 'black box.' The engineer sees numbers and graphs but loses a deep understanding of the process’s physics."
This isn’t a critique of computers per se. It’s a critique of methodological degradation, which occurs when:
In the late-Soviet school, LAF was a mandatory skill—on par with reading blueprints—because computers were scarce, and every engineer had to prove a result with a pencil. In the post-Soviet school (and even more so in today’s digital one), this skill has become elite. And now, an 87-year-old woman in 2026 is saying: without this skill, we design blindly.
The PDF on familysavelev.com is 17 MB, 65 pages of scans. Vision OCR managed to read the key sections.
This is important: the author didn’t "eyeball" anything—she calculated everything down to the millimeter.
S_б = U_б² / Z_в = 889·10⁶ VA — base (limit) power
S_ф = S_б / 3 = 296·10⁶ VA — per-phase power
U_б = U_н / √3 = 24·10³ V — stator voltage
t_б = 1 / (ω₀·r₀) = 3.18·10⁻³ s — base time unit
I_б = S_б / U_б = 1387·10³ A — stator current
R_б = U_б / I_б = 0.0171 Ω — stator resistance
Ψ_б = U_б·t_б = 4.41 Wb — stator flux linkage
L_б = Ψ_б / I_б = 2.06·10⁻³ H — stator inductance
I_f.ном = 1230 A — nominal excitation current
M_f = U_б / (ω₀·I_f.ном) = 50.6·10⁻³ H — mutual inductance
These aren’t random numbers. This is a full per-unit normalization of a three-phase machine with a nominal voltage of 24 kV (not 20 kV, as in most serial TVV units—the TVV-800-2U3 has a voltage of 24 kV, which results in a reduced stator current of 1387 A, critical for commutation). The base inductance of 2.06 mH is typical for machines of this class.
The author constructs a general structural diagram of the excitation system, which includes:
The manuscript contains separate diagrams for each block: amplifier (p. 15), current block (p. 16), thyristor converter (p. 17), frequency setpoint block (p. 18), voltage block (p. 19), reactive current sensor (p. 20). This is a full decomposition into 8 components, each with its own transfer function, which the author either takes from factory documentation or derives through transformation of the original circuit (p. 14—"Transformation of the TG Structural Diagram").
Here’s the meat of the method. The author takes the closed inner excitation regulation loop, writes its open-loop transfer function G_p = (1+τ) / (τ + p·0.03)², then the closed-loop one:
G_3 = (1+τ) / [1 + (1+τ)/(τ+p·0.03)²] = 1.4 / [p²·0.0036 + p·(0.06+2τ·0.06) + 1+1.4]
= 1.4 / [0.0036 p² + 0.233 p + 1]
Then, the Laplace transform for the transient process under a step input X_1(p) = 1/p:
F_1(p) = (1/p) · (2.33 / (p² + 0.233 p + 1)) = C_1/p + (C_2·p + C_3)/(p² + 0.233 p + 1)
And via the inverse Laplace transform, she obtains the original:
f(t) = 0.58 + 0.58·e^(-0.116 t)·sin(0.958 t - ε)
This is a classic example of the LAF method: instead of numerically solving a fifth-order characteristic equation, the author breaks the system down into elementary components, draws the asymptotic LAF for each (a segment with a −20 dB/dec slope), combines them graphically—and obtains the LAF of the entire system. From this LAF, you can immediately see whether the system is stable and what its phase and amplitude margins are. A computer does this in seconds, but the physical meaning is lost in the process. A LAF plot, however, is the physics of the process, written in pencil.
The author uses the simplified power transfer equation for a generator operating in a system:
P = (U₁·U₂ / X_d') · sin δ
where X_d' is the transient reactance (a key parameter for dynamics—it characterizes the generator’s behavior in the first seconds after a disturbance). She considers a short-circuit scenario in the system: the voltage U₂ at the busbars drops, and the power P drops proportionally. To prevent the generator from losing synchronism (i.e., to keep the rotor from "running away" relative to the stator field beyond the critical angle), the excitation must be increased as quickly as possible—i.e., the ceiling voltage must be applied to the rotor. After the short circuit is cleared, X_d' increases, and excitation must be increased again. The author shows that a properly tuned AVR-SDP1 provides 250% excitation forcing in ~0.05 s—and this is sufficient to maintain dynamic stability.
Here’s the amusing part—on this page, the calculation of the "economic effect from implementing the LAF method" follows the Soviet "Methodology for Determining the Economic Efficiency of Capital Investments" with a normative coefficient E_n = 0.15:
E = C₁ + E_n·K₁ − (C₂ + E_n·K₂)
The result? 7,500 rubles per year (in 1985 prices, which, through indexing, translates to roughly $100K–150K at today’s exchange rate, if you calculate very roughly). This is the direct economic effect from saving machine time and engineering hours by using the LAF method instead of numerical calculations. In 1991, this was a substantial sum—about 10 average monthly engineer salaries.
V_Savelieva is a new Habr user: "Invited today at 07:26 via invitation from @petuhoff" (plus or minus, this might be connected to knowing Pyotr, either through acquaintances or otherwise). 12 reputation points, 1 publication. Before this—complete silence. Viktorina Yuryevna Savelieva, an 87-year-old electrical engineer, independently digitized her 1991 manuscript (in her words—using "modern technologies"), uploaded the PDF and HTML to her family website (domain familysavelev.com), and today posted this work on Habr. This is a conscious act of passing down engineering heritage.
In the USSR, LAF was a basic skill in the curricula of TOE (Theoretical Foundations of Electrical Engineering) and AEP (Automated Electric Drive) departments at MPEI, LETI, UPI, and NETI. Every graduate plotted LAFs by hand at least 30–40 times during their studies. In post-Soviet Russia, this skill became niche—it survived among a narrow circle of specialists in automatic power system control but fell out of the general engineering curriculum. Today’s electrical engineering graduate knows how to plot a LAF in MATLAB but can’t sketch an asymptotic LAF with a pencil—because no one taught them how.
And now, an 87-year-old woman appears and says: "without this skill, we design blindly." This isn’t nostalgia for the USSR. It’s a diagnosis of the modern engineering school: we have powerful modeling tools, but we’ve lost the ability to read their output. And this isn’t just an energy sector problem—it’s a problem in any industry where a computer has become the intermediary between the engineer and the physics of the process.
I won’t pretend that the LAF method is a magical replacement for numerical modeling. It isn’t. LAF can’t:
But LAF can:
In modern engineering practice, the combination of numerical modeling (for precise answers) and frequency-domain methods (for understanding the physics) is the gold standard. Savelieva’s work is precisely about this combination.
There are several layers here, each interesting in its own right.
LAF is an analog way of thinking about systems. The method emerged before computers and is optimized for a person with a pencil. There’s an elegance to it: the method doesn’t require a machine for a first approximation. When an engineer sees a graph, they see the physics. When they see a massive array of 10⁶ numbers, they see only an array. And this isn’t a question of computing power—it’s a question of the interface between the engineer and the system.
In the USSR, an electrical engineer was both an artist and a mathematician: they drew graph paper, derived formulas in the margins, kept notebooks of plots. This isn’t an outdated skill—it’s a separate culture. And when 87-year-old Viktorina Yuryevna digitizes her 1991 work, she’s not just preserving a manuscript—she’s saving a way of thinking. It’s the same as if an old embroiderer uploaded her pattern diagrams online—what matters isn’t just the result, but the act of passing down the skill itself.
Surgutskaya GRES-2 is a symbol of the Soviet energy project of the 1980s. Eight 800 MW units, commissioned in 1985–1986. All of Western Siberia was powered by these machines. And in 1991, when the USSR was already collapsing, an engineer received an assignment from this station—to study the stability of its core. This is work done in the final days of the empire, and it has reached us as a manuscript. It’s an archival document of the era—like Gogol’s diary or Tsiolkovsky’s formulas.
Savelieva’s main thesis: "the LAF method is ideal as a supplement to primary computer modeling." This isn’t reactionary thinking—it’s a sound engineering position: use the computer for heavy lifting, but don’t forget to verify the result with physical meaning. And this is what’s missing in modern engineering education.
And the final layer is the act of publication itself. 87 years old, a lone author, a full manuscript, a link to a family website, an appearance on Habr via invitation. This is an act of courage. And this act deserves to be told—because tomorrow, people like her won’t be around anymore. And if we don’t digitize their legacy now, in 10 years, we’ll have an engineering school with computers but no language to speak to our teachers.
This isn’t a "story for the evening"—it’s a topic for a serious conversation in the engineering community. Here’s what can actually be done: