Allowed tolerance band

Required thickness or case depth. Indication of a tolerance band allowance... 

Fabrication tolerances can be equally important details. Normal tolerances for vessels are not adequate. The ASME tolerance for out of roundness of 1% is not always acceptable. Peaking and banding tolerances may need to be much lower than the Code allows. Offsets between plates should be carefully controlled. 

This will effectively create a tube surrounding all the calibration points. This approach requires optimisation of eqn (6.11), although constraints (6.10a) and (6.10b) are obviously not useful here. However this may be solved quite easily if we consider a value instead of unity. Such a value is denoted e and is called the width of the band or error tolerance or, simply, the 8-band. The conceptual idea implicit in e is very familiar to analytical chemists because if - by definition - the calibration points minimise their (overall) distance to the hyperplane, e represents no more than the residuals of the calibration fit. Thus, if we were to include all samples in the e-band, e would equal the largest difference between the target concentration of the samples and the values predicted by the SVR model. However, we know that some calibration points may be wrong and that random errors occur during calibration and the measurement of the unknowns and, therefore, it would be wise to allow the algorithm some flexibility, in the same way as was discussed for SVC. 

The above-proposed approaches allow the accuracy of external calibration to be approached without access to the sample position. The exact method of calibration selected will depend on the applications requirements and modeling tolerances (e.g. PLS for isomer-splitter or band-area integration for DLC measurements). 

---