Why Rockets Fail: The Few Root Causes Behind Most Launch Losses

Of the roughly 6,000 orbital launch attempts in history, about 5 to 6 percent have ended in failure. That number sounds small until you remember what sits on top: a satellite that took a decade to build, a crew, or in ISRO's early years, a nation's entire space ambitions riding on a single vehicle. The failures cluster. Pull apart any launch accident investigation report, from the PSLV-D1 anomaly in 1993 to the Antares explosion at Wallops in 2014, and the same handful of root categories appear, dressed in different technical clothes.

Turbopumps move propellant at extraordinary rates, the Space Shuttle Main Engine's turbopumps moved liquid hydrogen at roughly 70 litres per second, and they spin at tens of thousands of revolutions per minute under temperature swings that would destroy most metals. Cavitation, the formation and violent collapse of vapour bubbles inside the pump, is among the most common propulsion failure triggers. It degrades pump performance abruptly, and if the pressure drop is steep enough, the engine simply loses thrust at the wrong moment.

Combustion instability is a related threat. Propellants burning in a chamber can couple with acoustic resonance inside that chamber and create oscillating pressure waves that tear hardware apart in milliseconds. The F-1 engines on the Saturn V took years of testing and baffled injector redesigns to tame this problem. Seal failures and turbopump bearing failures account for a further share. In the Antares RD-181 predecessor failure in 2014, a turbopump oxygen-rich preburner was implicated. Propulsion failures account for roughly 40 percent of all launch vehicle losses across the orbital era.

A rocket ascending through the lower atmosphere is not in the vacuum its upper stages will eventually reach. Max-Q, maximum dynamic pressure, is the moment when aerodynamic stress on the vehicle peaks, typically between 10 and 15 kilometres altitude. The structure must survive it while already being stressed by internal tank pressure and the vibration of engines below.

ISRO's PSLV-D1, the first developmental flight in 1993, suffered a control system anomaly that led to attitude deviation and structural breakup at this phase. The Falcon 9 CRS-7 failure in 2015 traced to a strut inside the second-stage liquid oxygen tank, a component rated to withstand far more than the load it experienced, that broke and allowed a helium pressure vessel to escape. The strut failure was structural, but the root was quality control. These two failure modes, aerodynamic load exceedance and structural component failure, are often listed separately but share a common driver: margin that looked adequate on paper, tested against conditions that did not fully represent flight.

A rocket that produces full thrust but points the wrong direction is still a failure. Guidance, navigation, and control systems must integrate sensor data in real time and issue corrective commands through gimballing engines or reaction control thrusters. Software errors in this chain have caused several high-profile losses.

The Ariane 5 Flight 501 in 1996 is the canonical case: a 64-bit floating-point number representing horizontal velocity was converted to a 16-bit signed integer, overflowed, and triggered a cascade that ended in the vehicle self-destructing 37 seconds after launch. The software had been validated, for Ariane 4, which flew a different trajectory. The reuse of code without validating it for the new vehicle's flight envelope destroyed a rocket that had never misfired mechanically. ISRO's own software maturity across PSLV and GSLV generations is one reason the vehicles' reliability improved markedly between the 1990s and 2010s: each failure fed a tighter verification cycle.

A multi-stage rocket must shed mass at precise moments. Separation events are violent by design, explosive bolts, pyrotechnic cutters, and pneumatic pushers all fire in rapid sequence. A single pyrotechnic device that fires late, early, or not at all can cause the spent stage to collide with the live stage above it, or prevent the upper stage engine from igniting cleanly.

The GSLV Mk I D1 failure in 2001 involved the cryogenic upper stage, which failed to ignite after separation, a failure mode that combines stage separation sequencing with propulsion in a single critical window. Fairing separation failures, where the nose cone protecting the payload does not release, add mass that the upper stage cannot compensate for. The Vega-C failure in December 2022 traced to a nozzle throat insert on the Zefiro-40 second stage, but the investigation also examined the separation sequence upstream. These events happen in seconds. There is no opportunity to intervene.

Some failures are not failures of hardware or software but of process. A component installed with the wrong torque. A sensor reading that was flagged and dismissed. A test waiver that removed a margin that turned out to matter. The Columbia accident, though not a launch failure, is the clearest case: foam strikes on the orbiter had been observed on previous missions and reclassified as acceptable. The reclassification was wrong.

Range safety systems, the flight termination systems that destroy a vehicle if it deviates toward populated areas, are a separate layer that occasionally becomes the proximate cause of a loss when they activate on a vehicle that might have recovered. The distinction between a vehicle that failed and a vehicle that was terminated is important for investigators but irrelevant to the payload.

What the failure record across NASA, ISRO, ESA, Roscosmos, and SpaceX shows is that the categories are stable even as the technology changes. Propulsion, structure, guidance, separation, and process account for the overwhelming majority of losses. Reliability improves not by eliminating the categories but by narrowing the margins within them, through redundancy, testing fidelity, and the willingness to treat an anomaly as a signal rather than an outlier. Chandrayaan-3's successful lunar landing in 2023, after Chandrayaan-2's lander failure in 2019, is a precise illustration of that cycle working as intended.