Pea and lentil flour quality as affected by roller milling configuration

This study examined the effects of roller mill configuration on pulse flour quality. Dehulled yellow pea and green lentil were ground to flour using a laboratory roller mill characterized by its flexibility to control particle size reduction while maintaining a constant feed rate. The milling diagram length (long, six passes vs. short, four passes) and sieve sizes (large, 300 μm vs. tight, 180/150 μm) were adjusted for a total of four milling configurations. Each flour stream was characterized with respect to its physical properties and chemical composition. No notable differences were identified between pea and lentil based on how the milling configuration influenced flour characteristics. Overall, combining streams to produce a whole flour did not affect the chemical composition but resulted in variability for physical characteristics as indicated by a tendency toward increased levels of damaged starch with the shorter milling diagram. Damaged starch content was found to be indirectly associated (p < 0.05) with the particle size distribution, where the highest concentrations were noted in flours with median distributions below 30 μm. When individual streams were compared across milling configurations, the stream itself was rarely found to significantly influence flour physicochemical properties. However, the variation exhibited in particle‐size distribution, protein, starch, ash, and damaged starch content could have practical relevance given the many significant (p < 0.05) correlations with functional properties that could subsequently affect the end‐use applicability of flours. This would imply that specialized flours could be made with the intention of being used for defined food applications.

configurations. Each flour stream was characterized with respect to its physical properties and chemical composition. No notable differences were identified between pea and lentil based on how the milling configuration influenced flour characteristics. Overall, combining streams to produce a whole flour did not affect the chemical composition but resulted in variability for physical characteristics as indicated by a tendency toward increased levels of damaged starch with the shorter milling diagram. Damaged starch content was found to be indirectly associated (p < 0.05) with the particle size distribution, where the highest concentrations were noted in flours with median distributions below 30 μm. When individual streams were compared across milling configurations, the stream itself was rarely found to significantly influence flour physicochemical properties. However, the variation exhibited in particle-size distribution, protein, starch, ash, and damaged starch content could have practical relevance given the many significant (p < 0.05) correlations with functional properties that could subsequently affect the enduse applicability of flours. This would imply that specialized flours could be made with the intention of being used for defined food applications.

| INTRODUCTION
The world population is increasing steadily, and food supply will be a major challenge for the 21st century (Chéreau et al., 2016). Plantbased foods provide an opportunity to address this challenge; it follows that pulses will play an important role in food market evolution as the world progresses toward more sustainable food sources (McDermott & Wyatt, 2017). Pulses are high in protein having an amino acid profile complementary to that of cereals and contribute to environmental sustainability through their low carbon and water footprints and their importance in crop rotations (McDermott & Wyatt, 2017;Todorov et al., 1996). The Food and Agriculture Organization (FAO) projects global pulse consumption to increase by 21 million tons over the period 2013-2025 (Rawal & Navarro, 2019); 68% of this increase in consumption will be in the form of food (Rawal & Navarro, 2019). In order to grow the pulse industry on a global level, a greater understanding of ingredient, that is, flour, specifications for targeted end-use applications, is required, as is the case with cereal flour production (Campbell, 2007;Prabhasankar et al., 2000). The ultimate aim of the pulse flour industry is to advance knowledge on how best to adjust the milling process in order to deliver flours that are of a desired and consistent quality despite annual and batch-to-batch variations in raw material quality and composition.
A literature review compiled in 2018 outlined the status of pulse flour milling research (Scanlon et al., 2018). This review identified that although studies have demonstrated acceptable pulse flour use in a range of applications, there is limited information on how to manipulate pulse milling to produce flours optimized for specific food product applications. The authors summarized research gaps that need to be addressed in order to obtain the same level of technical expertise that exists in the wheat milling industry (Campbell, 2007). Several research gaps are directly related to pulse flour processing. Is there a desired particle size and how to produce it? What are the target flour specifications and the corresponding by-products? What is the influence of starch damage and how to monitor this characteristic?
Multiple technologies, including impact, hammer, roller, stone, and attrition milling, have been shown to be effective in producing pulse flours (Maskus et al., 2016). Within studies on each of these processes, there was large variation in sample origin, preparation, and milling configuration. It follows that the milling technology itself and the applied process settings, as well as differences in the raw material due to seed variety or growth environment, will have an effect on flour properties, which might be attributed to changes in the particlesize distribution (PSD) and level of starch damage (Maskus et al., 2016). Variability in these flour properties will have an effect on the respective end-use applications (Scanlon et al., 2018).
Unlike other milling technologies, roller milling facilitates the production of a diversity of flours (Campbell, 2007;Prabhasankar et al., 2000;Sakhare et al., 2014) Also, a roller mill requires lower specific energy consumption compared with other technologies such as pin and attrition mills. Several studies have investigated the milling of pulse products using roller mills (Mittal et al., 2012;Sakhare et al., 2014;Sakhare et al., 2015;Thompson et al., 1976;Zinn et al., 2002). The production of a specific flour using roller mills can be performed more gradually as a result of the multiple roll and sifter passes. It follows that many adjustments to the milling process are possible, providing greater control over particle-size reduction and generation of flours with a variety of functional characteristics. This has created interest in the use of roller milling technology to control variability and manage diversity in pulse flour quality.
All streams from any given milling process may yield flours of variable composition and quality (Prabhasankar et al., 2000). Additionally, the roller milling of pulses will be affected by many parameters including dehulling, tempering, roll gap, sieve openings, milling diagram length, fluting profile, roll differential speed, and feed rate. The aim of this study was to assess the utility of roller milling for the production of pulse flours with quality characteristics that might be optimal for particular applications. The study was designed to highlight whether variations in processing parameters, specifically pulse type (pea or lentil), milling diagram length, and sieve openings, were sufficient to produce flours differing significantly in their physical and chemical characteristics and, therefore, in end-use quality.
Dehulled samples of pea and lentil were milled using a laboratoryscale roller mill with variations in two processing parameters, namely, milling diagram length and sieve openings. The effect of mill flow configuration on the physical and chemical properties of the flour streams produced was analyzed. A subsequent study will relate differences in flour quality to end-use applicability.

| Raw material
Samples of yellow pea (CDC Spectrum) and green lentil (CDC Greenstar) harvested in 2019 were sourced from a producer in Limerick, Saskatchewan, Canada. Whole pea (24.4% protein on dry basis, 10.7% moisture) was dehulled and split with a stone mill (MTI 70, Saint donat sur l'herbasse, Moulins Alma Pro®) spaced at 6 mm, and then passed through a zig-zag classifier (MZM 1-40, Augsburg, Germany, Hosokawa-Alpine®) set at 3.6 m/s. Lentil (24.8% protein on dry basis, 9% moisture) was dehulled by a combination of a knife mill (SM300, Haan, Germany, Retsch®) set at 1200 rpm equipped with a 6-mm sieve and the zig-zag classifier set at 3.6 m/s. Nondehulled seed was collected with a seed cleaner (RKS SLN, Kitzingen, Germany, Pfeiffer®) and returned to the dehulling process. The dehulled seed of pea and lentil represented 89.9% and 85.0%, respectively, of the initial sample mass; with 8.9% and 14.6% collected as hull and 1.2% and 0.4% lost during the process (dust and humidity). After dehulling, the residual hull on the seed was manually sorted as 0.3% for lentil and 0.6% for pea. The residual hull material was collected as a by-product stream from the milling process.

| Roller milling
A number of parameters that may be adjusted during the roller milling process have an effect on final product quality, including milling diagram length, roll fluting profiles, speed differential, sieve openings, and stream recycling (Campbell & Bedford, 1992). The aim of this study was to produce flours differing in composition and quality by adjusting two of these process parameters, namely, milling diagram length and sieve openings.
The roll gaps and upper sieve openings for the three corrugated break rolls (B1-B3) and the three smooth reduction rolls (C1-C3) were as follows: B1 with a 400-μm gap and 710-μm upper sieve, B2 with a 100-μm gap and 600-μm upper sieve, B3 with a 60-μm gap and 530-μm upper sieve, C1 with a 60-μm gap and 530-μm upper sieve, C2 with a 50-μm gap and 400-μm upper sieve, and C3 with a 30-μm gap and no upper sieve. Milling was conducted using 36-kg samples at a constant feed rate (6 kg/h) with temperature controlled to <30 C to ensure that no flour heating occurred.
These four diagrams produced 52 streams, among which 40 were flours and 12 were by-products composed mostly of hull.

| Analysis
The 40 flours were characterized in terms of their yield and chemical, physical, and functional properties. Most analyses were done in duplicate (only density was done in triplicate). Nitrogen content was determined by the Kjeldahl method (TecatorTM with Kjeltec 8400, Hillerød, Denmark, Foss® system) according to procedure NF EN ISO 5983-2 and using a nitrogen to protein conversion factor of 6.25.
Moisture and ash content were measured using a Prepash system (Dietikon, Switzerland, Precisa®) according to Sluiter et al. (2008). The test consisted of drying at 105 C until a constant weight was achieved, followed by calcination at 550 C. Total starch and damaged starch contents were determined by enzymatic methods using Methods for evaluation of functional properties were adapted from protocols previously described in the literature (Maskus et al., 2016). Viscosity evaluations were performed on a Rapid Visco Analyzer RVA 3D + (Watham, United States, PerkinElmer, Inc.). Four grams (3.6-g dry matter) of sample was dispersed in 25 ml of deionized water. The following temperature and mixing profile was applied: t 0 à 0 min, 50 C and 960 rpm/t 1 à 0.1 min, 50 C and 160 rpm/t 2 à 1 min, 50 C and 160 rpm/t 3 à 4.42 min, 95 C and 160 rpm/t 4 à 7.12 min, 95 C and 160 rpm/t 5 à 11 min, 50 C and 160 rpm. Three viscosity parameters were calculated from the resulting curve, namely, maximum viscosity, the hold on trough (obtained after total dispersion and thus the loss of granular structure just before cooling) and the final viscosity (obtained at the end of cooling).
Foaming properties were evaluated with a Foamscan (Lyon, France, Teclis Scientific) using a 0.1% protein solution. Foam was formed by bubbling air into the solution using a frit (P3) at a flow rate of 200 ml/min for 30 s. Foam volume and its stability were then recorded over 10 min, using egg white as a reference.
Emulsifying properties of flours were measured by producing an oil-in-water emulsion. A dispersion of 1% flour in water was mixed with oil at a proportion of 75/25 (dispersion/oil) followed by sonication at room temperature with a Branson 450 Digital Sonifier (Connecticut, USA, Branson Ultrasonics ©) using a 13-mm probe with 10 cycles of 10 s over 2 min set to 40% magnitude. Size distribution of the oil droplets was then measured on a particle-size analyzer (Mastersizer; Malvern, United Kingdom, Malvern Panalytical) with two dispersants (water and SDS). A refractive index of 1.46 was used for sunflower oil and 1.33 for water, and an absorption index of 0.01 was used for sunflower oil. Emulsion stability was assessed by comparison of initial droplet size distribution and after 7 days (percentage difference).
Water holding capacity was measured by adding flour to 10 ml of water at a concentration of 20 mg/ml of dry matter. Suspensions were mixed for 1 h with stirring, followed by centrifugation at 15,000g for 10 min. Water holding capacity was expressed as the ratio of the water content of the hydrated sample and the original sample weight.
Powder bulk density was evaluated using a 250-ml graduated cylinder. Aerated bulk density was measured as the density obtained immediately after adding the flour to the graduated cylinder, and tapped bulk density as the value obtained after 1 min of manual vibration. Hausner index, indicative of flowability, was calculated as follows (Thakur et al., 2019): where ρ t = tapped density and ρ a = aerated density. A Hausner index below 1.2 was considered to indicate high flowability.

| Streams calculation
Blending of flour streams is a common practice during roller milling as the chemical composition of flours follows additive laws. The concentration of a particular parameter for a particular flour combination was calculated as follows: • Y i is the concentration of the selected parameter (ex: protein, damaged starch, particle size, etc.) • x i is the percentage of flour in the blend Entire streams were blended in order to avoid having an infinite number of flour combinations.

| Statistical analysis
Mean differences among streams were assessed using Tukey-Kramer groupings for parameters identified as significant with Type III analysis of variance at a significance level of p < 0.05. Relationships among flour properties were assessed using Pearson's correlation coefficient.

| Stream production
As presented in Figure 1, dehulled pea and lentil were milled using four milling configurations obtained by adjusting the milling diagram length and flour sieve sizes, thereby producing a total of 40 flour streams.
Analysis of variance (ANOVA) results demonstrated that moisture and damaged starch content differed significantly (p < 0.05) among streams, whereas no effect was found for all remaining physicochemical properties (results not shown). Despite the lack of significance in some samples, differences among all streams will be discussed in order to gauge their practical relevance. The composition and yield for each stream are presented in Tables 1 and 2 for pea and lentil, respectively.
T A B L E 1 Yield, composition, and physicochemical properties of roller-milled pea flour streams The sum of the yields for all flour streams of a given milling configuration did not equal 100% due to the production of by-products.
However, the by-product yield was always less than 0.5%, with the exceptions of middlings (fine by-product) produced from pea using the long milling diagram/tight opening (1.8%) and bran (coarse byproduct) produced from lentil using the short milling diagram/tight opening (3.6%). The higher amounts of by-products for these configurations were attributed to an overload of certain sieves due to the constant feed rate applied. The 20 streams produced for each pulse type (pea or lentil) exhibited large differences in yield. However, pulse type itself did not have a large effect on the amount of flour obtained for a particular stream. Therefore, for both pea and lentil, the majority of flour was derived from the C1 stream, followed by B1, B2, and occasionally C2. It follows that B3, C3, and usually C2 were low yielding streams (Tables 1 and 2). These differences in stream yields were expected, because they are well known in cereals (Campbell, 2007;Prabhasankar & Haridas Rao, 2001) and in pulses (Sakhare et al., 2014(Sakhare et al., , 2015. Ash, starch, and protein content of flour streams ranged from 2.5% to 3.2% and 2.6% to 3.1%, 45.5% to 59.7% and 41.9% to 57.8%, and 23.3% to 28.1% and 23.6% to 29.5% for pea and lentil, respectively, demonstrating moderate differences among streams (Tables 1 and 2). Although streams were not found to be significantly different in composition (p < 0.05), except for moisture and damaged starch content, the variation in protein content among streams may have practical relevance as pea and lentil having initial seed protein concentrations of 24.4% and 24.8% (dry basis), respectively, produced flour streams differing in protein concentration by up to 5% (Tables 1 and 2). Blending of flour streams is a commercial strategy commonly used to produce wheat flours with desired physicochemical properties (Campbell, 2007;Mittal et al., 2012;Sakhare et al., 2014). The roller milling process thus is a useful tool for adjusting the protein content of pulse flours and providing opportunities to optimize protein requirements for particular end-uses.
Damaged starch content displayed significantly (p < 0.05) higher values for streams collected at C3 (xc = 11.9%) than at B2 (xc = 3.5%) for both pea and lentil. Similar results were reported by Price et al. (2021), where the final fourth middlings stream produced from roller milling of pea displayed significantly (p < 0.05) higher levels of starch damage than all other streams. Variability in flour PSD was large (CV > 50%); however, no significant differences were found among streams. Particle-shape distribution was assessed for the majority of samples; however, because no significant differences in particle shape were detected, the data are not presented.
T A B L E 2 Yield, composition, and physicochemical properties of roller-milled lentil flour streams   (Pelgrom et al., 2013). In this process, the protein-rich fractions are found to be lower in starch and higher in ash. This is explained by the disintegration of the cotyledon and the concomitant release of fine particles (<10 μm) rich in protein, whereas starch granules and/or agglomerates of protein and starch constitute fractions with particles larger than 10 μm (Moller et al., 2021). In this study, no correlation was found between protein content or starch content and particle size ( Table 3).

| Links between process parameters and flour quality
Production of protein-rich streams was thus a consequence of the roller milling configuration that led generally to particle-size reduction without a substantial change in particle composition. Table 3 shows that for both pea and lentil, a clear negative correlation existed between particle size and damaged starch content (p < 0.01). This indicates that a flour with a smaller average particle size will exhibit a higher level of damaged starch. The same relationship was found for stone milling (Monnet et al., 2019) and is well known for rice and wheat milling (Thakur et al., 2019). Figure 2 demonstrates that this relationship is not linear but rather is characterized by a threshold near an average particle size of 35-40 μm, below which the damaged starch content increases markedly to above 10% (on a starch basis). For coarser flours (above 40 μm), the level of damaged starch averaged below 8%, with levels closer to 4% for flours with an average particle size over 100 μm. In this study, the average particle size was utilized as an aggregate value to study the effect of particle size. Conclusions were similar when considering other criteria such as d 90 (Tables 1 and 2 Table 4, although the differences are not large and may be attributed to differences in the composition of the samples milled. PCA of flour physicochemical characteristics was selected to further emphasize the relationships between pea and lentil flour parameters given the multicollinearity of the data.    viscosity data were used in the correlation analysis. Also, results for emulsification capacity in water or in SDS were strongly correlated (r > 0.87, p < 0.001); hence, data for emulsification in water only are reflected in Table 3.
Emulsion capacity and foam volume were weakly correlated with any of the physicochemical parameters analyzed. Therefore, variability in values for these properties cannot be explained by the present study. Table 3 demonstrates the independency of most of the functional properties analyzed as they correlate weakly with one another.
The majority of the exceptions are for those parameters measured during the same test, such as foam stability and foam volume, which were positively correlated. The same observation is valid for tapped density, aerated density, and Hausner index; all three of which also were correlated to the RVA trough and final viscosities.
Of the correlations presented in Table 3, those found to be significant (p < 0.05) for both pea and lentil flour streams include the following: foam stability with damaged starch content; final viscosity with protein, ash, starch, and damaged starch content; aerated density with damaged starch content; average particle size and Hausner index with damaged starch content; and average particle size and SPAN.
The final viscosity was inversely related to damaged starch, ash, and protein content but directly related to starch content and mean particle size. Due to the absence of a correlation between damaged starch (itself correlated with median particle size) and protein content (which was correlated with starch content and ash content), pasting properties appear to be affected by both physical and chemical parameters. This multiparameter response indicates the complexity with which all physicochemical properties of a flour can affect its functionality.
Aerated density and Hausner index, both relevant to flour handling, were significantly affected by particle size and damaged starch content. Due to the relationship between particle size and damaged starch, it may be assumed that particle size is the main explanation for this observation. Results indicate that if a higher flour density is desired, too fine of a powder should be avoided to control the level of starch damage. It follows that flowability (Hausner index < 1.2) of a flour is better when the particle size is relatively coarse (>60 μm) and narrowly spread (low SPAN).
Damaged starch content also was found to be significantly correlated with foam stability, although the relationship between the parameters is complex (Figure 4). It appears that if the damaged starch content is below 10%, the foam is not stable. Above 10%, improved stability is observed, until above 12% starch damage where the foam becomes stable. This behavior is hypothesized to be the result of the increased viscosity that arises alongside a certain degree of damaged starch that is required to stabilize the foam. However, three flour streams did not follow this trend as they exhibited foam stability with starch damage below 8% (Figure 4). These three flours were derived from the B3 and C3 streams, both of which were low yielding under every milling configuration. One explanation could be that these low yielding streams were contaminated with other compounds/particles (such as hull) that imparted increased viscosity useful in stabilizing the foams. The B3 and C3 streams correspond to the final flour collection before by-product removal.

| CONCLUSION
The objective of this study was to assess the flexibility of roller milling for the production of flours with specific quality characteristics that may be useful for particular end-use applications. The results fully demonstrated that a variety of flour streams could be produced using the roller mill configurations applied. Streams varied with respect to their particle size and protein, starch, and damaged starch contents.
The potential to create flours with varying physicochemical characteristics could be even higher at industrial scale, considering that this study focused on only two process parameters (milling diagram length and sieve openings), whereas commercially, many additional factors can be altered to influence final product quality. Further research is required to examine the influence of these additional milling process parameters and how they may affect the applications of a flour.
F I G U R E 4 Correlations between damaged starch content and foam stability for roller-milled pea (white points) and lentil (black points) flour streams and combinations of flour streams Despite the high degree of variability among flour streams produced, results from this study showed that combining all flour streams from any given milling configuration will produce whole flours having similar chemical characteristics. The physicochemical characteristics of a particular flour stream across milling configurations were alike for both pea and lentil and therefore, may also apply to the milling of other pulse types using the same settings. A shortened milling diagram with tighter sieve openings increased the severity of the mechanical treatment during flour production, resulting in a tendency for increased levels of damaged starch. For the production of flours with lower levels of starch damage, a longer and therefore more gradual milling diagram with larger sieve openings should be applied. An alternative approach is the blending of similar flour streams. For example, the second break (B2) roll streams exhibited lower levels of damaged starch and could be combined to produce flours with low starch damage and higher pasting properties, whereas the final reduction (C3) roll streams could be mixed to produce flours with relatively high damaged starch contents. This study also showed that stream blending would be a useful strategy for creating flours that meet particular end-use specifications. The next step in this research is the undertaking of food application trials designed to help develop ideal flour specifications.
Finally, this study demonstrated that the physicochemical properties of streams translated to differences in the functional characteristics of the flours. The behavior of the functional properties of a flour was complex, with only a few parameters being clearly related to the physicochemical characteristics studied. In particular, pasting properties were favorable when both protein and damaged starch contents were low. Additionally, foam stability increased with a higher level of damaged starch, and flowability was improved for flours found not to be too fine (>60 μm) and having narrow SPAN. Food application studies are required to aid in understanding the effect of these functional attributes and how they relate to the final product. Considering the inherent flexibility of roller milling for producing flour streams with particular characteristics, food incorporation studies are very important to highlight ideal flour specifications for a given application.

CONFLICT OF INTEREST
Authors JC and TD are employees of Pulse Canada.

FUNDING INFORMATION
This study was supported by Pulse Canada.