Can you provide some more detail? Are the lots in chronological order (time series) or are they just sorted from high to low failure rates? I suspect it is the former in which case you can try to fit time series models to this data. On the other hand, just a graphical exploration may be sufficient. I find that a simple power function fits this data quite well (failure rate = 57813*time^-0.668). Graphically, assuming this is a time series, the failure rate appears to be stabilizing and the long-run prediction from that power function might be appropriate - but I'd need more information to know if that makes any sense).
