Proving a Stirling engine can survive launch, and then keep running for 17 years

What the work claims

This is a hardware status paper from the NASA lab that leads Stirling radioisotope power system work, not a paper announcing a new result with a single headline number. Its claims are cumulative. First, longevity: Technology Demonstration Convertors TDC #13, #15, and #16 have each surpassed 17 years of continuous extended operation and hold world records for the longest-running flexure-bearing free-piston Stirling engines, while ASC-0 #3 has almost reached 14 years as the record holder among gas-bearing designs and ASC-L has passed 9 years.¹ Second, robustness: the newest Sunpower Robust Stirling Convertor, SRSC #4, has completed all four tracks of the Dynamic Radioisotope Power Systems verification and validation plan (baseline performance, random vibration, static acceleration, thermal cycling) with no sign of permanent degradation beyond sub-watt power shifts.¹ Third, system integration: the Stirling Generator Testbed, four convertors around one radiantly-coupled heat source, produced a measured 16.8% conversion efficiency at its first quasi-steady baseline operating point on May 31, 2023, which the authors contrast with the 6.3% efficiency of current radioisotope thermoelectric generators.¹

How it works

A free-piston Stirling convertor is a heat engine with one moving part: a piston that oscillates inside a sealed helium-filled pressure vessel, shuttling gas between a hot end and a cold end through a regenerator, driving a linear alternator. It converts heat to electricity at potentially two to four times the efficiency of thermoelectric couples, which trade efficiency for absolute reliability because they have no moving parts. The entire reliability question for a Stirling radioisotope power system therefore reduces to demonstrating that the moving machinery survives launch environments and runs for a mission lifetime, and that is precisely what the Stirling Research Laboratory is built to test.¹

The design-life accounting matters for mission planners: the 17-year target is not arbitrary. It is 14 years of active mission duty plus 3 years of pre-launch storage, and because the only degradation mechanism the convertors exhibit is predictable material creep in the hot pressure vessel, accelerated testing cannot shortcut the calendar. Predictive failure modeling works for failure modes you can provoke quickly; here it cannot, so the lab simply keeps convertors running, unattended, year-round, and has been doing so since the early 2000s.¹

The qualification campaign on SRSC #4 is the paper's engineering core. Track 2 exposed the convertor to qualification-level random vibration at the Structural Dynamics Laboratory: a peak of 7.7 g_rms calculated for a 75 kg generator mass assumption under the GEVS standard, in three orthogonal orientations. Output power wobbled between about 37.3 W and 38.1 W during full-level vibration, then settled; the permanent shift after all three axes was -1.0 W for SRSC #4 (SRSC #3 shifted +0.6 W), judged a pass against measurement uncertainty. Track 3 took the convertor to Case Western Reserve's geotechnical centrifuge in December 2023: a 2-hour dwell at 6.3 g in four lateral orientations (power dipped less than 2 W and recovered), then ramps to 22.5 g held for 5 seconds, during which the convertor was deliberately stalled to avoid an over-test condition, since the centrifuge decelerates more slowly than the required profile. Permanent shift: -0.4 W. Track 4 ran 13 thermal cycles from April 2024, each about 8 hours, between a hot condition of 720 degrees C hot-end (nominal plus 20 degrees C margin) and a cold condition with the hot end at 16.5 degrees C, over about a month. Permanent shift: -0.2 W.¹

Two details keep these numbers honest. The lab re-quantified its own instrumentation: convertor electrical power is measured to plus or minus 2.0 W, temperature to plus or minus 3 degrees C, so every "pass" above rests on the ability to repeat the operating point, not on resolving a fraction of a watt absolutely. And the sensitivity study behind the predictions is empirical: 39 operating points (31 unique) on a hermetic SRSC #3, perturbing piston amplitude, frequency, and three temperatures, regressed into a linear model that reproduces measured alternator power well enough to serve as an in-rack anomaly detector.¹

The generator testbed is the system-level step. Four convertors of two heritage designs (two hermetic gas-bearing Sunpower ASC-E3, two non-hermetic flexure-bearing Infinia SRG110 engineering units) sit in one housing around a centrally located, radiantly coupled 500 W electric heat source. Pair-level redundancy lets each pair throttle to match load, and the first attended 24-hour run at roughly 500 degrees C hot-end produced the 16.8% efficiency figure. The paper is explicit that this point was not optimized for efficiency, and that the genuinely system-level demonstrations (heat loss through insulation and housing, failure of a convertor pair, throttle-up of the survivors) are still future work.¹

Steelman: the evidence strategy is exactly right

Everything here is measured hardware in named environments, and the measurement boundaries are stated. The 16.8% is a bench measurement of a real four-convertor assembly, not a projection; the vibration profile is a published standard (GEVS, GSFC-STD-70008) at qualification level; the acceleration levels map to launch, reentry, and spin-stabilization cases; the thermal cycle bounds include margin explicitly quantified (plus 20 degrees C, minus 15 degrees C). The longevity claim is uniquely strong because it does not depend on models at all: three machines have simply run for 17 years, which for a technology whose central doubt is wear-out is close to dispositive. The paired-convertor architecture also answers the right system question, which is not "how efficient is one convertor" but "what happens to power output when one of four dies," and designing the throttle-up answer into the housing from the start is sound fault-tolerant practice.¹

Where a skeptic should push

The most load-bearing assumption is that component-level qualification plus long extended operation adds up to flight readiness, and the paper itself supplies the counter-evidence. Nothing here has flown as a Stirling radioisotope generator in this configuration; the SRSC is described as a "flight-like" design, the heat source in the testbed is electric, not radioisotope, and the fault-tolerance behavior that justifies the four-convertor layout has not yet been demonstrated, only scheduled. Second, the efficiency comparison invites misuse: 16.8% from an unoptimized baseline at 500 W heat input is a legitimate measurement, but comparing it to 6.3% for a flown, fueled, thermally integrated RTG compares a testbed point with a system, and the paper's own "3 to 4 times more efficient than conventional thermoelectric RPS" phrasing (citing prior program work) is a potential figure, not this paper's result. Third, the sub-watt permanence claims sit at or below the stated plus or minus 2.0 W instrument accuracy; the pass criterion is repeatability of the operating point, which is defensible but weaker than independent metrology. Fourth, all robustness results come from single units (SRSC #3 and #4); there is no statistical basis for a reliability claim yet.¹

What a 16.8% testbed says about RPS mass budgets

Implication: the deepest insight is about how reliability evidence is bought. For thermoelectric RTGs, lifetime assurance was nearly free because there was nothing to wear out; for dynamic conversion, the failure physics (creep in the hot pressure vessel) is slow, predictable, and unacceleratable, so the only way to retire risk is calendar time. The SRL's 20-plus-year accumulation of extended-operation hours is therefore not a testing program but an annuity: every year of unattended operation on TDC #13 through SRSC #4 buys down the same uncertainty, and a mission that wants Stirling-class efficiency in the 2030s is already relying on starts made in the early 2000s. Anyone planning a dynamic-conversion mission today should read the 17-year design-life target (14 years duty plus 3 years storage) as a schedule constraint on mission formulation, not a spec sheet line.¹

Threat: the specific-power gap is where hype will creep in. At 6.3% efficiency, a radioisotope source delivers about 63 W_e per kilowatt of heat; at 16.8% bench efficiency the same heat source would deliver over 2.5 times the electric power, which is why program literature projects two to four times the system specific power of thermoelectric RPS. But quoting that ratio as available today would launder a non-optimized, unfueled, electric-heated testbed point into a capability. The paper's own open items are the tell: heat loss through the housing, paired-convertor failure, and survivor throttle-up are all still ahead, and each one erodes the system number. Mission advocates who budget Stirling specific power before those tests close are pricing hardware that has not demonstrated its fault modes.¹

Opportunity: the paired-convertor throttle architecture is the genuinely transferable design pattern. Most power sources fail gracefully or not at all; here, the four-convertor housing with a central radiant source is explicitly built so that losing two machines halves output rather than ending the mission, and load-following by throttling convertor pairs replaces the bang-bang shunt regulation of classic RTGs. For lunar surface and outer-planets missions where load varies across day-night cycles and instrument duty, that mechanism (each pair throttled up or down to meet specific power requirements, in the paper's words) converts redundancy from parasitic mass into an operating mode. Combined with the SAGE-model regression used as an in-rack anomaly detector, it is a blueprint for power systems that report their own degradation rather than concealing it.¹

The bottom line

Separating what is demonstrated from what is projected: demonstrated, with named hardware and environments, are multi-decade convertor life, qualification-level robustness of a flight-like Stirling convertor with sub-watt permanent shifts, and a 16.8% measured system conversion efficiency at a single unoptimized baseline point. Projected, and not yet shown, are fueled-system efficiency, fault-tolerant throttle-up behavior, and any flight of this configuration. Confidence is high that Stirling convertors are mechanically robust and long-lived at bench scale; moderate that the four-convertor generator will meet its efficiency potential once insulated, fueled, and fault-tested; and low that any of this translates to specific-power claims before the planned system tests close. What would confirm the claim: a fueled generator testbed run with measured heat-source-to-bus efficiency and a demonstrated convertor-pair failure and recovery. What would break it: the pending paired-convertor failure test showing throttle-up cannot regulate the shared thermal load.