An Approach for Obtaining Stable, Reproducible, and Accurate Fibrogram Measurements from High Volume Instruments

Abstract

Fiber length is a crucial property throughout many textile processing steps. The high volume instrument (HVI) is the most commonly used tool to assess the properties of cotton fibers. The HVI currently reports two length parameters, the upper-half mean length (UHML) and uniformity index (UI). The UI is the ratio of mean length (ML) to the UHML expressed as a percentage. UHML and ML are extracted from the fibrogram. These parameters are used in the current U.S. cotton classification and global cotton marketing systems. The two values are highly correlated and characterize only the longer fibers in a sample. The fibrogram holds more descriptive information than the two measurements provided by the HVI. However, limited information is available about the stability and repeatability of the fibrogram measurement. This study aims to investigate the stability of the fibrogram, assess the reproducibility across multiple instruments, and determine if corrective actions are required. Three different raw cotton sample sets were tested for this three-stage experiment. The obtained results demonstrate that for a given HVI, the entire fibrogram is stable over both the short term and long term; however, differences among HVIs were observed. The proposed correction procedure effectively reduces the differences among the four HVI lines.

1. Introduction

The quality of cotton and its production depends on many factors such as the varieties grown, production methods, environmental conditions, handling, and mechanical processing [1,2,3]. Fiber length is one of the most critical properties of cotton to producers and spinners [4,5,6]. The significance of this property in cotton marketing and processing has been reported by many other scholars [7,8,9,10]. Cotton producers are rewarded with a premium for their cotton with longer fibers and penalized with discounts for cotton with shorter fibers [11,12,13]. The information on the fiber length distribution can be used to adjust machine settings throughout the many processing steps found in spinning mills to translate fibers into yarn [14,15,16].

The two most common instruments used to assess cotton fiber length are the advanced fiber information system (Uster AFIS pro 2, Uster Technologies AG, Switzerland) and the high volume instrument (Uster HVI 1000, Uster Technologies AG, Switzerland). The AFIS is an individual fiber tester that provides a fiber length distribution among other fiber quality parameters [17,18,19]. To prepare a sample for testing, a 500 mg sample of raw cotton is used to make a 30 cm long hand-formed sliver. Then, the sliver is placed into the instrument feeding system. Cotton fibers are individualized through the inbuilt opening system and different fiber parameters are measured by sensors and reported. The AFIS is the most common system that provides a complete fiber length distribution; however, its throughput is low, making it costly and impractical for large commercial and research applications.

The HVI is the most-used tool by the cotton industry to assess and classify cotton fiber quality. It is a bundle fiber-testing instrument and its length measurements are based on the fibrogram theory [20,21]. The fibrogram is a graphic representation of the length distribution of the fibers in the cotton beard. Suh and Sasser explained how an HVI operates to measure different properties of cotton fibers [22], including the length measurements described here. To test a sample, about 10 g of raw cotton is placed in a fibrosampler (Figure 1a). The HVI comb picks fibers at random and forms a fiber beard. The fiber beard is then brushed to remove loose fibers and extraneous matter (Figure 1b) and moved to a sensor to be scanned with a beam of light (Figure 1c).

The HVI length measurement begins scanning the beard at 3.81 mm away from the comb [23]. As scanning begins near the comb, the maximum amount of light is attenuated by the fiber beard, which is used to normalize the measurement to 100%. When scanning continues toward the tip of the fiber beard, the amount of light attenuated by the beard continues to decrease as fewer fibers are available to be scanned. Finally, it reaches a point where there are no fibers available to be scanned, resulting in 0% attenuated light. Chu and Riley assumed that the amount of light attenuated decreases proportionally with the number of fibers being scanned [21]. The graphical representation of the distance traveled by the fiber beard versus the attenuated light is called the fibrogram, which the HVI uses to measure fiber length (Figure 2).

Currently HVI derives two length parameters from the fibrogram: upper-half mean length (UHML) and mean length (ML). However, it reports the UHML and uniformity index (UI). The UI is the ratio of ML to UHML expressed as a percentage. These reported length parameters are used in the current U.S. cotton classification and global cotton marketing systems. The ML is expected to be the average length of all fibers and the UHML is the mean length by number of the upper-half longer fibers by weight [24]. Sayeed et al. reported that the UHML and ML are closely related to the 1.8% and 7.8% span lengths extracted from a fibrogram, respectively, and only characterize the longer fibers as shown in Figure 2 [25]. They also demonstrated that the whole fibrogram holds useful fiber length information that is not captured by the currently reported length parameters and that these new data improved the prediction of yarn quality. While it is promising, there is no information on whether the fibrogram is a stable measurement. In addition, the current HVI outputs are calibrated measurements while the fibrogram is not.

Figure 2. An illustration of a typical fibrogram and the current HVI measurements extracted from it [25].

Figure 2. An illustration of a typical fibrogram and the current HVI measurements extracted from it [25].

It is common in the cotton industry to refer to HVI outputs as “calibrated” measurements. However, examination of the HVI “calibration” method reveals that it is based on a software correction and not a mechanical calibration of the instrument. The HVI uses slopes and offset values of cotton fiber samples with a known length value to correct the length parameter measurements. Therefore, we use the term “corrected” rather than “calibrated” because the proposed approach is software based and does not require instrument adjustment.

This study aims to assess the stability of the fibrogram over both short-term and long-term periods, evaluate the reproducibility across multiple instruments, and determine if a correction procedure is needed.

2. Materials and Methods

Three independent raw cotton sample sets were used for a three-stage experiment. The first set was to investigate the stability of the fibrogram measurements for two periods (short term and long term) on one HVI, the second set was to evaluate the agreement among HVIs, and the third set was to devise a correction procedure for the fibrograms. Samples were conditioned at 21 ± 1 °C and 65 ± 2% RH for at least 48 h before testing.

2.1. Experiment 1: Fibrogram Stability Study

To identify potential samples to be used in the fibrogram short-term and long-term stability studies, 72 samples were grown in 2016 across the Texas High Plains. They are “commercial-like samples” as they meet most of the requirements of commercial cotton production. They are commercial varieties grown, harvested, and ginned like commercial cotton. They consist of 12 commercial varieties grown in five different locations on a large scale. A stripper harvester was used to harvest the cotton from these locations. In addition, these samples were ginned with a commercial-type gin. These samples were tested with the HVI using a standard testing research protocol (10 replications of the fibrogram). Then, a subset of six samples covering the widest possible range of fiber length variation captured by the fibrogram was selected. The six samples were measured using a standard testing research protocol of 10 replications with the length/strength module testing mode [26] over two different periods of time. This mode allowed for testing of length and strength only and export the fibrograms after testing. A series of 36 tests were performed over a two-week period for the short-term stability evaluation and 15 tests for a total of eight weeks for the long-term stability assessment. The same USTER HVI 1000 at the Texas Tech University (TTU) Fiber and Biopolymer Research Institute (FBRI) was used for both stability assessments.

The HVI length/strength module testing mode provides the functionality to save the reported data in a Microsoft Excel file. Within that file, the fibrogram for each replication is stored as a vector image from which fibrogram data can be extracted as a series of 81 data points using a MATLAB script [25]. If the fibrogram is transposed and linearly interpolated, length (scanned distance) will be the response variable. Using this representation, 99 lengths are interpolated in increments of 1% occluded light-producing span lengths from 1% to 99%. The span length indicates a corresponding scanned distance or length of fibers in the beard for a specific fraction of fibers assuming the presence of 100% fibers at the starting point [27,28]. The average of 10 replications at each percent span length was used to compute a representative fibrogram for each sample. Then, a set of eight span lengths (2%, 8%, 15%, 25%, 40%, 50%, 60%, and 75%) along the entire fibrogram curve representing short, average, and long fibers were chosen for analysis.

We looked at scatter plots (length vs. time) for multiple fibrogram measurements taken over the two periods to detect potential trends. To assess the variability of the measurements, we calculated the coefficient of variation (CV %) of each span length from each sample over each time period.

2.2. Experiment 2: Assessment of the Fibrogram Reproducibility among Instruments

The United States Department of Agriculture (USDA) Agricultural Marketing Service (AMS) retests 1% of the samples tested in classing offices around the nation. From these retested samples, we were able to obtain 3210 samples for this study. These samples from commercial bales represent diverse cotton varieties grown in 2018 at different locations across 13 states. These samples were tested for HVI and AFIS fiber properties. The HVI data (UHML and UI) and the AFIS data (upper quartile length (UQL) and short fiber content by number (SFCn %)) were used to select a subset of 333 samples, which covers the widest possible range of fiber length variation. Then these 333 samples were measured using a standard testing research protocol of 10 replications using the HVI length/strength module testing mode [26] on three USTER HVI 1000s at the TTU FBRI. Fibrograms were retrieved in the same manner as described in the previous experiment.

The fibrograms from the 10 replications were averaged to produce a representative fibrogram for each sample and on each separate HVI. These representative fibrograms were then compared across the three instruments by using paired t-test on the eight chosen span lengths from each fibrogram. Linear discriminant analysis (LDA) was used as a tool to model fibrogram variances among the HVIs and to summarize the total variations.

2.3. Experiment 3: Devising a Correction Procedure

The HVI corrects the data extracted from the fibrogram (ML and UHML) and then reports two corrected length parameters (UHML and UI). On the other hand, exported fibrograms are not corrected; therefore, an investigation was needed to determine if a method like the current HVI correction method can be applied to bring the fibrogram measurements to a similar level across multiple instruments. Currently, the USDA-AMS creates two standard calibration cottons (long/strong and short/weak) and generates their reference values for the UHML and UI following an ASTM standard protocol [29]. During the correction procedure, these two samples were tested with 12 replications to obtain the observed values for the UHML and ML. Simple linear regression was performed between the average of 12 replications and the reference values separately for UHML and ML. Then, the slopes and the offsets of the regression equations were used to correct the measurements. The reported UI by the HVI was calculated using the corrected UHML and ML.

We followed a similar procedure for correcting the fibrogram; instead of correcting UHML (1.8% span length) and ML (7.8% span length), we corrected eight span lengths across the fibrogram. Since there is no reference measurement available for the fibrogram, the same two USDA length calibration cotton standards were used to establish reference values for the fibrogram. The two cottons were tested with 100 replications using the HVI length/strength module testing mode on two HVIs (USTER HVI 1000s) at the TTU FBRI, providing 200 fibrograms for each cotton. The averages of these 200 fibrograms for each calibration cotton were considered the “reference” fibrograms. These samples were also tested on four other HVIs (USTER HVI 1000s) with 12 replications using the same testing mode. The average of the 12 fibrograms for each cotton and each HVI were considered “observed” fibrograms.

Simple linear regressions between the reference and the observed values were used as a means to obtain correction factors (slopes and offsets) for a series of span lengths (2%, 8%, 15%, 25%, 40%, 50%, 60%, and 75%). It is important to note that for each of these span lengths, we generated four correction equations—one for each HVI. Then, a set of 13 USDA evaluation samples that cover a wide range of fiber length parameters was selected. These samples were tested with 10 replications on the same four HVI lines (USTER HVI 1000s) within same day.

The uncorrected span lengths of each sample on individual replications were corrected using the correction equations previously determined. After the correction, the average of 10 replications at each percent span length for each HVI line was calculated to get a representative fibrogram for that sample. The values at each percent span length were then compared across the four instruments before and after the correction.

3. Results and Discussion

3.1. Fibrogram Stability Study

The stability of the length measurements at each percent span length for each of the six samples over time was evaluated. The average fiber length measurements of 10 replications for a given percent span length was computed and plotted.

3.1.1. Short-Term Visual and Trends Assessment

Scatter plots of length measurements at several span lengths versus time (two weeks) are shown in Figure 3 as an example to visualize potential trends. The middle portion of the fibrogram (i.e., 25% SL, 40% SL, 50% SL, and 60% SL) for all samples are shown as an example and to visualize the short-term potential trends. The six samples are presented to illustrate the stability of the measurements for a wide range of length variations.

The scatter plots in Figure 3 do not show any visible upward or downward trend for the length measurements. To confirm this, a simple linear regression at each span length for each sample was performed on a series of 36 tests measured over a two-week period. The slope and offset values were used to assess the presence of trends, which could indicate machine drift. For data with no drift, the offset is the fiber length measured at the given percent of occluded light while the slope should be zero. The best fit line should be horizontal with the measured values scattered randomly around it. Our results for the short-term assessment revealed that the slopes are not significantly different from zero at a 95% confidence interval, indicating no significant upward or downward trend in the measurement over time.

3.1.2. Variability of Selected Span Lengths over the Short Term

Regression analysis was used to assess the presence of trends for the short-term period and the coefficient of variation (CV%) was used as a standardized method for measuring the spread of the measurements. For each sample, the CV%s among 36 tests (over time) were calculated for each percent span length for the short term. The coefficients of variation ranged from 0.49% to 3.67% for the short-term study (Figure 4).

An increase in CV% from the longer span lengths (lower percentages) to the shorter span lengths (higher percentages) could be due to a decrease in the signal-to-noise ratio of the sensor. The greatest occlusion of light occurs at the base of the beard, which means very little light, i.e., signal, reaches the sensors, thereby allowing electrical noise to have a greater influence on the output. As it was reported, the length measurement variations of the fibrogram at each span length for the short-term period is small (i.e., CV% below 5%) and acceptable. As explained earlier, some variation in fiber properties is expected because of the inherent nature of cotton (indeterminate growth, growing conditions, harvesting, ginning, etc.).

3.1.3. Long-Term Visual and Trends Assessment

The long-term stability assessment has 15 observations distributed over a period of eight weeks. The stability of the length measurements at each percent span length for each sample over this period was evaluated. The average fiber length measurement of 10 replications for a given percentage of fibers (% span length) was computed and plotted against time. In the same way, the scatter plots of length measurements at each span length for each sample for the middle portion of a fibrogram for all samples are shown in Figure 5 as an example to visualize potential long-term trends.

The multiple length measurements for the long-term assessment are shown in Figure 5. Interestingly, no trends were observed. This is confirmed with simple linear regressions between length values at each specific span length and time. As discussed earlier, slope and offset values can be used to assess the presence of trends. The results of the long-term assessment show that the slopes are not significantly different from zero at a 95% confidence interval, indicating no significant upward or downward trend in the measurement over time.

3.1.4. Variability of Selected Span Lengths over the Long Term

For each sample, the CV% among 15 tests (over time) were calculated for each percent span length for the long term. The coefficients of variation among tests range from 0.73% to 4.35% in the long-term period (Figure 6). As described in the previous section, the signal-to-noise ratio of the sensor may decrease for the longer fibers during scanning of the fiber beard, increasing the CV% from the lower-percentage span lengths to the higher-percentage span lengths.

It can also be observed that the variation in the fibrogram length measurements at each span length is small (i.e., CV% below 5%) and acceptable in the long-term study. For that reason, the fibrogram measurements appear to be stable throughout the study period.

3.2. Assessment of the Fibrogram Reproducibility among Instruments

3.2.1. Fibrogram Measurements across HVIs

The 10 replications per test measured on each HVI were averaged at each span length over the fibrogram curve to obtain representative fibrogram for each sample. Figure 7 illustrates the variation among HVIs for one sample.

In Figure 7, visible differences in the fibrograms are observed among instruments. These differences are concentrated in the fibrogram portion between 10% and 70% SL. Similar differences are observed across HVIs for all samples. As the extracted fibrograms are raw, uncorrected data, differences among instruments are expected. We performed a paired t-test on eight representative span lengths to assess the statistical significance of the observed differences between HVIs. The results for all possible combinations of HVIs are shown in

Table 1. Summary of a paired t-test (α = 0.05) of length measurements between HVIs for all possible combinations of HVIs.

Paired HVIs	Parameters	Percent Span Lengths (% SLs)
		2%	8%	15%	25%	40%	50%	60%	75%
HVI 1 and HVI 2	t Stat	16.55	3.19	3.18	2.20	−2.83	−8.19	−19.98	−7.97
	P(T <= t) two-tail	<0.001 *	0.002 *	0.002 *	0.029 *	0.005 *	<0.001 *	<0.001 *	<0.001 *
HVI 1 and HVI 3	t Stat	11.42	23.64	23.64	28.73	31.89	30.93	16.47	−10.52
	P(T <= t) two-tail	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *
HVI 2 and HVI 3	t Stat	−4.88	20.03	20.03	26.05	34.89	40.85	38.89	-1.54
	P(T <= t) two-tail	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *	<0.001 *	0.125

* Differences are statistically significant at α = 0.05.

. Our statistical analysis showed that the differences in length measurements between the HVIs at each span length are statistically significant at the 0.05 level.

When we consider the differences between HVIs for the 1.8% and 7.8% SLs, although they are statistically significant at the 95% confidence interval, these differences are not of practical importance. They are within acceptable tolerances for the HVI [30]. We know that the 1.8% SL and 7.8% SL represent the regular HVI outputs. It is of interest to note that these values (1.8% SL and 7.8% SL) are very close, even if not corrected. However, there is a need to develop a fibrogram correction procedure for most span lengths to bring different HVIs to the same level.

3.2.2. Modeling Fibrogram Fiber LENGTH Measurement Variances across HVIs

The total variations in fiber length captured by the fibrogram have been modeled using linear discriminant analysis to assess the reproducibility of fibrogram measurements among instruments. The analysis was conducted by assigning the 99 span lengths across the fibrogram as covariates and the HVIs as categorical variables. As shown in Figure 8, fibrogram data are clustered for each HVI, except for one outlier sample from HVI 1 appearing near the cluster for HVI 3. HVI 1 is separated from HVIs 2 and 3 based on the canonical axis 1. In addition, HVI 1 and HVI 2 fibrograms appeared different based on canonical axis 2.

The entropy RSquare ranges from 0 to 1 and indicates highly interconnected groups (0) to completely distinct groups (1). Therefore, the higher the value for this quantity, the better the quality of the classification or separation. The results showed a clear separation with an entropy RSquare of 0.98. This outcome is consistent with the paired t-test results reported in the previous section that showed statistically significant differences among HVIs. Overall, our results show that fibrograms vary from HVI to HVI. Therefore, there is a need to develop a fibrogram correction method to bring the measurements from different HVIs to the same level.

3.3. Devising a Correction Procedure for the Fibrogram

3.3.1. Establishing the Correction Procedure

Following the correction protocol outlined in the Materials and Methods section, the slopes and offsets for each HVI line were computed based on their corresponding observed values and reference values for a series of span lengths as shown in

Table 2. Linear correction parameters for length data extracted from the reference cotton fibrograms.

HVIs	Parameters	Percent Span Lengths (% SLs)
		2%	8%	15%	25%	40%	50%	60%	75%
HVI 1	Slope	0.998	1.014	1.020	1.031	1.028	0.988	0.901	1.094
	Offset [mm]	−0.274	−0.713	−0.924	−1.052	−0.947	−0.492	0.214	−0.421
HVI 2	Slope	1.005	0.968	0.939	0.920	0.908	0.877	0.781	0.705
	Offset [mm]	−0.651	−0.160	0.112	0.160	−0.029	0.095	0.641	0.753
HVI 3	Slope	1.015	0.986	0.975	0.951	0.926	0.904	0.876	0.841
	Offset [mm]	−0.815	−0.026	−0.160	0.126	0.201	0.295	0.301	0.153
HVI 4	Slope	1.100	1.067	1.060	1.074	1.078	1.075	1.032	1.230
	Offset [mm]	−2.931	−1.632	−1.204	−1.004	−0.613	−0.368	0.034	−0.519

. The slopes and offsets were used as linear correction factors for the individual replications of each sample for each HVI line.

3.3.2. Testing the Correction Procedure

The length values among HVIs are at different levels for different span lengths, and the level of variation among HVIs varies from span length to span length. The four HVIs have noticeable differences near the middle portion of the fibrogram, particularly at 25% SL, 40% SL, 50% SL, and 60% SL before correction (previously reported). For illustration, the average length values of two span lengths (25% SL and 60% SL) are plotted for the 13 samples before and after correction (Figure 9 and Figure 10, respectively). For each figure, part (a) is before correction, and part (b) is after correction. It is noticeable from these figures that the correction procedure brings the fibrogram measurements closer across the four HVIs.

The coefficients of variation between HVIs were calculated for each sample and each span length to determine whether the proposed correction method minimizes the variation among HVI lines. In most instances, the coefficients of variation (CV %) among HVIs decrease after applying the correction equations (Figure 11 and Figure 12). The correction procedure reduced the CV% to an acceptable level, i.e., below 5%.

We also investigated whether differences among HVIs are reduced after correction for a series of 6 HVI pairs (HVIs 1–2, 1–3, 1–4, 2–3, 2–4, and 3–4) using the same 13 samples. Figure 13 shows the average differences in lengths across the HVI pairs.

As shown in Figure 13, the correction procedure effectively reduces the global difference between HVI measurements. For the span lengths 25%, 40%, 50%, 60%, and 75%, the differences before and after correction are statistically significant with a 95% confidence level.

4. Conclusions

We investigated and analyzed the stability and reproducibility of multiple fibrogram length values obtained within and between the high volume instruments (HVIs). Stability studies showed good stability of the fibrogram, and we observed acceptable levels of variation among measurements. The results revealed that the fibrogram length values obtained from the HVI were stable over both short-term and long-term periods. Yet, the results observed on multiple instruments showed that the fibrograms for the same sample vary from HVI to HVI. Variation between HVIs was more concentrated in the central part of the fibrogram, and these differences were found to be statistically significant at the 0.05 level.

The proposed correction procedure effectively reduces the level of differences among the four HVI lines. In the span lengths near the middle portion of the fibrogram where there are noticeable differences, the differences before and after correction are statistically significant with a 95% confidence level. In addition, the HVIs are considerably closer to the acceptable limit after correction (e.g., with a few exceptions, CVs are below 5%).

Recent studies using the complete information contained in the fibrogram suggested its potential importance to the cotton industry. However, there is a need for a fibrogram correction method to bring different HVIs to the same level and identify the salient features of the fibrogram for use in the cotton industry. This result could also help identify important span lengths extracted from a fibrogram that better capture the variability within the sample than the length parameters currently reported. Although it is a promising result, it needs to be validated on a larger set of commercial samples.