Abstract :
The paper demonstrates that it is impossible to discriminate, even between widely disparate batches, by the application of a constant numerical value in the form of a ?factor of safety? in conjunction with a ?guaranteed minimum? testing technique. It may even be that the batch with the poorer characteristics has by far the better chance of acceptance under such circumstances. The introduction of a new set of empirical standards is recommended to replace existing ?factors of safety?. The new standards should be ?guaranteed minima,? upon which, by statistical methods of appraisement, great reliance might be placed. The statistical methods should be so devised that they substantially eliminate the necessity of empirical allowances to cover unknown degrees of variations in batch characteristics, which cannot be reliably forecast at present from results obtained from a small sample out of the batch. Thus we should eliminate the rough-and-ready technique of asking a ?guaranteed minimum? strength of 10000 lb. for line insulators while that for conductors carrying exactly the same load is only 8000 lb. The reason for the difference is that the minimum cannot actually be guaranteed, but a greater range of variation is probable in insulators than in conductors. By statistical methods, however, the minimum can be very substantially guaranteed. The limits of error in statistical technique are examined in some detail and the results are compared with those for present methods of appraisement, to the disadvantage of the latter. The advantages of statistical control over the processes of production and the setting of acceptance levels which may be expected with a reasonable degree of confidence to be related to the actual batch characteristics are summarized at the end of the paper. It is also shown that present ideas of an adequate size of test sample are parsimonious to the extent of ineffectually, and it is recommended that test samples of at least 50 units should become common- ly accepted as reasonable.
