A simpler criterion would now be more valid, either one simply related to the cladding melting temperature in the range of 2500°F, or a strain due to thermal stresses (say 0.5 or 1% strain) at 1700 to 1800°F, or the boiling point of the coolant sodium (1632°F at atmospheric pressure). The reason­ing for the latter (5) is that although sodium boiling does not infer cladding rupture, conditions are probably such that cladding failure is not far away.

3.1.1.1 Accident Severity Levels

It is now possible to define accident severity levels for gas-bonded oxide fuel in stainless-steel cladding. Table 3.1 shows typical severity levels for

start- and end-of-life conditions. These may be used in a classification of core accidents (Section 3.1.3). The actual values will depend critically on the fuel pin design.

3.1.2 Steam Generator Failure

Other components of the plant can fail; one of the more probable being the steam generator in which a sodium-water interaction becomes a possi­bility in the LMFBR system.

Table 3.2 shows the modes, the locations, and the causes of failures experienced to date, some of the failures being primary and some secondary as noted. Such a survey of nuclear facilities gives some indications of practical experience. Out of 16 facilities, only 6 reported failures, and all of these were due to faults that could have been caught by adequate quality assurance and inspection techniques, rather than being due to behavior outside normal operating conditions. The locations of the failures were mainly in the tube sheet or at the tube-to-tube sheet joint.

From this information a failure rate can be calculated for sodium and nonsodium steam generators: (a) sodium operation, 21 failures per 10® tube-hr; and (b) non-sodium operation, 29 failures per 10® tube-hr (6). Thus the experience with sodium is not significantly different from the experience in water-to-steam systems. This failure rate would imply a failure about every 3 or 4 years for a 700 MWt plant.

This information can be used to assess availability and maintainability and can perhaps be used to establish some quantitative failure probabilities for fault-tree analysis. However such information cannot be used for establishing the point of steam generator failure during abnormal op­eration.

Similar failure rates can be established for all other components in a nuclear power plant and this is currently being organized in the US by the Liquid Metals Engineering Center (7). The center collects all failure data available for liquid-metal systems and analyzes the incidents for cause and origin. Naturally, statistics are still poor, but Table 3.3 shows some of the results available. Note how misleading the failure rate for diesel generators might be since it is as yet derived from only four failures. The low failure rate for core fuel and breeder elements is worth noting because almost all the data arise from fuel failures in the Enrico Fermi plant.

The accumulation of failure data is a necessary part of safety engineering. At present it should be considered as long-term data which will be available when fault-tree analyses for plant failures become quantitative sometime in the future.