*2.9. Evaluation of Linearity, Limit of Detection and Lower Limit of Quantification (Detectability), Intra-Day and Inter-Day Repeatability (Precision), and Recovery Test (Trueness)*

Solutions with increasing concentrations of diluted pure standards were prepared to evaluate the linearity of the instrumental response. The range of concentrations (0.02–1.00 mg/L) was established on the basis of the data already present in the literature and by exploring the chromatographic traces of some real samples. Because biscuits can cause a matrix effect that can affect acrylamide determination, linearity was also evaluated using an analyte-free biscuit matrix in which known concentrations of acrylamide were added.

In the aqueous reference solution, the instrumental limit of detection (LOD) and the lower limit of quantification (LLOQ) were obtained by applying the equation:

$$\text{LOD or LLOQ} = (\text{K} \times \text{sy}/\text{x})/\text{b} \tag{3}$$

where sy/x and b are the estimated regression standard deviation and the slope of the relevant analytical calibration function, respectively. K = 3 and K = 10 were chosen in order to obtain the LOD and LLOQ, respectively.

Precision was evaluated with an intra-day repeatability test on a sample and a standard solution of acrylamide (0.125 mg/L) each injected five times, and an inter-day repeatability test carried out on a sample injected over the course of five consecutive days in the same conditions. The relative standard deviations were calculated for each substance.

The trueness was evaluated through a recovery test where a known quantity of a standard acrylamide solution (0.25 mg/L) was added in the presence of the internal standard in samples consisting of a biscuit matrix previously deprived of the analyte and applying the extraction and chromatographic determination protocols.

#### *2.10. Statistical Analysis*

Employing different aliquots of each sample, three repetitions of each measurement were performed for each kind of analytical determination. Data were expressed as mean values (±standard deviations). FDOE was set using Microsoft Excel 365.

Because parametric ANOVA assumption tests (such as normality, equal variance and equal or near-equal sample size) were not satisfied, we performed the non-parametric Kruskal–Wallis equality-of-population rank test [27] and the non-parametric Wilcoxon rank-sum (Mann–Whitney U test) [28], using the temperature program (TP), the steam release time (SRT), and the ammonium bicarbonate percentage (AB) as statistical factors to observe the differences in median values in the sample set. Nevertheless, the ANOVA was also run, and its results were compared to those of the Kruskal–Wallis test.

An evaluation of the significant correlation among all parameters that showed significant differences in the Wilcoxon rank-sum (Mann–Whitney U test), i.e., thickness, L\*, a\*, b\*, *aw*, and acrylamide concentration, was performed using the non-parametric Spearman rank order correlation test to assess correlations among acrylamide and the other variables. All tests were performed using Stata/SE 11.0 for Windows (StataCorp LP, College Station, TX, USA).

### **3. Results and Discussion**

#### *3.1. Optimization of the Reference Industrial Biscuit (RIB)*

The use of industrial-scale machinery for the preparation of the SIB implies substantial differences compared with the samples obtained using laboratory-scale equipment. For this reason, preliminary tests were carried out to try to balance the most evident differences and obtain experimental samples with specific physical characteristics, such as thickness, *aw*, and color similar to those of the SIB.

The challenge of this research work lay in obtaining good reproducibility of the industrial process conditions in producing the RIB with lab equipment while simultaneously trying to emulate the characteristics of the SIB. Therefore, the first step was to analyze the differences between the two systems in detail to understand the causes of divergence

between the RIB and the SIB. Some technological differences were, however, deemed insurmountable. One such difference was the ability to develop the same extent of gluten network during the creaming phase owing to the higher quantity of industrial heat that commonly triggers a more relevant leavening effect in the biscuits. Other technological differences, such as molding and the speed of production, were considered less relevant.

The differences between the two baking systems may have a tremendous impact on the experimental biscuits. After several attempts, a temperature program that made RIB almost identical to the SIB was finally found (Figure 2):

**Figure 2.** Experimental samples 4 (**a**) and 6 (**b**) and reference biscuit (RIB) (**c**).

160 ◦C (2 min) − 170 ◦C (3 min) − 140 ◦C (3 min).

However, the resulting biscuits had *aw* lower than that of the SIB and a slightly lower height, whereas the color was similar. Unfortunately, further variations of the TP led to a further negative impact on one of the measured parameters (*aw*), although height and color were very close to the original results. For this reason, it was decided to keep the height and color of the RIB as similar as possible to the SIB, to the detriment of the *aw*.
