Home Physical Sciences Calculation and prediction of thread consumption in technical textile products using the neural network method
Article Open Access

Calculation and prediction of thread consumption in technical textile products using the neural network method

  • Thouraya Hamdi EMAIL logo , Faouzi Khedher , Mohamed Jmali and Boubaker Jaouachi
Published/Copyright: December 29, 2025
Download Article (PDF)
AUTEX Research Journal
From the journal AUTEX Research Journal Volume 25 Issue 1

Abstract

This paper describes a method developed to improve the accuracy of thread consumption predictions for technical textile industrialists, particularly for products such as the automotive leatherette cover, by applying a neural network method. The approach relies on analysing a database regarding different fabrics, garment designs, and relevant production parameters to develop a suitable prediction model. The developed model using advanced supervised learning techniques and statistical methodologies, including multiple linear regression, captures the intricate relationships between many parameters to predict thread consumption accurately. The results show that the model without interaction achieves an R2 of 0.76, while the model with linear regression including interactions achieves an R2 of 0.91. In contrast, the use of neural networks achieves an R2 of 0.94. This framework is advantageous for manufacturers, as it allows them to optimize the organization of their production schedules and other operations. By integrating predictive analytics, manufacturers can source and manage raw materials more accurately, avoiding excess or insufficient production output, while improving productivity. Meanwhile, reducing the waste of materials helps in cost-reduction strategies and improves the financial efficacy of the business. This practice becomes sustainable by addressing the environmental concerns due to overproduction, one of the modern issues. In addition, accurately predicting thread consumption makes it easier to control the stock level of inventories, which is important for effective supply chain management. With optimized inventory, companies will save on storage expenses and minimize the downtime between operations, resulting in increased productivity. The combined application of neural nets and statistical techniques increases the accuracy of forecasts, which is essential to manufacturers in a highly competitive environment.

Keywords: yarn consumption; textile technical products; neural network

1 Introduction

The textile industry constantly tries to improve its processes and minimize costs, while being competitive internationally [1]. In order to achieve these goals, well-structured work systems should be put in place throughout the industry. Sewing threads are among the most important parts of the complex assortment of materials needed. Determining accurately the amount of yarn consumed for a particular order is very important for accurately estimating a company’s needs. This is, however, not an easy estimate to make as rough estimates ordinarily lead to too much stock being stored, which only compounds costs [2]. In order to control excess stock and stave off a shortage of materials that would paralyze production work, it is necessary to estimate sewing thread consumption for some selected models long before mass production begins [3]. As prices increase for raw materials in textiles, such as sewing threads, more attention has to be directed towards managing thread consumption [4]. Sadly, many clothing manufacturers base their calculations on production experience, which may not be accurate and almost always leads to overstocking, creating a wasteful supply chain with ineffectively managed threads [5].

Many research studies have investigated using neural network models to estimate and predict yarn consumption [6]. These models consider the history of clothing products, including details of fabrics, designs, and production settings, to create accurate consumption models. The use of neural networks is warranted, because they capture intricate patterns and dependencies, thus increasing the accuracy of yarn consumption calculations. Other researchers have used statistical methods to study the determinants of yarn consumption in garments. Regression and time series analyses have been used to study the interplay between fabric attributes, garment size, and production activities [7]. These statistical methods improve the estimates of thread consumption by identifying key parameters and formulating mathematical equations.

In another study, researchers analyzed sewing thread usage in jeans with fuzzy logic and regression methods to enhance the prediction accuracy within factors like thread mix, needle dimensions, and fabric weight [8]. Prior studies highlight a need to forecast yarn consumption before the mass production cycle begins to avoid surplus stocks versus a lack of stocks later [9]. By looking into different factors that impact the consumption of sewing threads, researchers have created models based on experimental and statistical data that provide straightforward estimations regarding various kinds of fabric [10]. Different studies have tried to compare geometric estimation approaches based on stitch forms using statistical methods such as multi-linear regression and other forms to improve estimation [11]. Different basic stitches have been analyzed concerning sewing thread consumption using images and other statistical methods. Results showed that image analysis has higher accuracy [12]. In the other study, the researcher used a neural network approach to estimate the sewing thread usage on jean trousers and compared the results with the regression models [13]. Studies concentrating on cover stitches have revealed that the most important factors affecting thread consumption are stitch density and seam width. These studies have indicated that geometric prediction methods are more accurate than statistical ones [14]. A few other studies focused on the influence of fabric compressibility, feed rate, thread tension, and sewing lockstitch shape on the amount of thread consumed [15].

The goal is to create models using artificial neural networks and regression techniques to estimate thread consumption in multilayered seam assemblies stitched by a lock stitch 301 [16]. Other studies focused on the correlation of sewing thread size, stitch density, and seam quality in woollen fabrics using available data science methods for analysis [17]. Models predicting thread consumption for lockstitch seams have included fabric and thread characteristics [18]. It was reported that the sewing thread consumption of stitch class 301 was accurately modelled using the Fourier series and image analysis [19]. A composite regression model was also developed to estimate thread consumption with cotton and polyester threads to simplify control in the garment industry [20].

The current developments in textile engineering through AI applications demonstrate multiple important research findings. The research by Malami et al. [21] demonstrated how machine learning models now predict fabric fiber-reinforced cement compressive strength as AI applications expand into material performance predictions. Dau Sy et al. [22] demonstrated how advanced optical and deep learning techniques enable automated fabric defect classification to improve textile manufacturing quality control through AI applications. Shanmughan and Kandasubramanian [23] examined multiple AI prediction methods for silk fibroin behavior to demonstrate how AI optimizes textile properties. The research shown the current advancements, while proving the importance of AI approaches for solving textile industry problems.

The research study demonstrates that current literature lacks essential information about using advanced AI methods for technical textiles particularly automotive leatherette covers. The current literature lacks comprehensive models which study the distinctive features and usage patterns of sewing threads in specialized products. The research targets these areas to create strong predictive models which combine linear regression with neural networks.

The purpose of this research is to identify the characteristics and types of sewing threads and to highlight their significant place as an essential element in the textile industry. The novelty of this paper lies in the fact that the materials selected for the study were derived from technical textile products. In this work, experimental and statistical analysis is performed to factor that influences sewing thread consumption and model equations are developed to predict consumption rates for different process parameters. The modeling is done in a gradual manner starting from linear regression without interactions, then moving to factor interactions, and finally looking for a neural model that can predict well with a high correlation coefficient. These issues can be addressed and the need for precise thread consumption quantities will be emphasised in order to reduce costs, improve inventory control and take ecological factors into consideration in the textile industry. If these estimated values and proper thread management policies are implemented by clothing manufacturers, they will be in a better position to compete in the global market and contribute to ecological sustainability.

2 Materials and methods

2.1 Computational tools and hardware specifications used

Matlab was chosen as the computational software for the study, because it enables the execution of complex calculations plus the modelling of several variables, especially sewing thread consumption for various materials and conditions. The laptop used for the study was Intel Core i7 powered with 32 GB DDR4 RAM, which offered seamless multitasking and flawless performance during complex simulations. It also came with a 512 GB SSD, which aided in the storage of data and quick retrieval when needed. To aid in the demanding tasks associated with this research, the laptop was also equipped with an NVIDIA GeForce GTX 1650 graphics card, which greatly enhanced the computational power and visual rendering capabilities.

2.2 Method of calculating sewing thread consumption

The thread consumption calculation involves a premeasured approach, which guarantees sharp precision. First, a thread length - often set at 1 m-is pulled and marked on the upper and lower spools before the sewing operation begins. This exercise ensures that handy data can be relied upon during evaluations. After the stitching has been completed, the remaining thread length is recorded, and the difference between the two set marks is assessed. This process indicates how much stitching thread was utilized in constructing the seam. This technique makes it possible to record and monitor the stitching thread consumption with great accuracy. It significantly helps evaluate sewing productivity and efficient resource allocation in different sewing operations. With this approach, many manufacturers can now understand thread consumption for different orders and, as a result, improve their planning and management and control costs. In addition, the practice contributes to more effective resource management by reducing material wastage, while enhancing the strength of sewing operations.

2.3 Factors affecting sewing thread consumption

Stitch Geometry: The complexity of designs leads to increased thread usage because they contain more details in both upper and lower stitches.

Stitch Density: The number of stitches per centimeter directly affects thread usage because each stitch requires a particular amount of thread.

Seam Width: The distance between needles or presser foot width determines thread consumption while thicker materials need additional thread to secure stitches.

Thread Size: The size of the thread determines how much material it uses for each stitch which affects the total thread consumption.

Spacing: The distance between needles or seam width determines the total thread length required for a project because wider spaces need additional thread.

2.4 Multiple linear regression

Multiple Linear Regression is a complex technique that employs two or more independent variables compared to a single dependent variable for modeling and analyzing relations deep within the data. Multiple linear regression adopts simple linear regression methods, where only one independent variable is considered, but shifts towards evaluating the effects of multiple variables simultaneously. The main focus of multiple linear regression is to estimate the best-fitting linear functional relation between the dependent variable and independent variables involved and to determine the impact of their respective changes on the dependent variable. This type of modeling has great relevance in many dimensions, such as social science, economics, engineering, etc.; as described in Equation (1):

(1) Y = β 0 + β 1 X 1 + β 2 X 2 + … + β n X n  

Where:

Y is the dependent variable.

β 0 is the intercept.

β 1, β 2, …, β n are the coefficients of the independent variables.

X 1, X 2, …, X n X n are the independent variables.

For efficient application of multiple linear regression, at least one of the following conditions must be met: Deviations from the observations should have a predictable linear correspondence to variations within the independent variable set by the model. The second factor, independence, states that the average correlation of sample observations taken from a population is zero. Also, across all independent variable values, there is no change in the degree of variance present among the errors, which is known as Homoscedasticity.

Finally, the residuals must follow a normal distribution to sensible reasoning; the normality of residuals should be checked from the model, because the inference here is from the residuals obtained.

Multiple linear regression serves two primary purposes: it predicts by calculating the value of the dependent variable owing to the independent variables, and it analyzes the relationships by looking at the influence of independent variables on a dependent variable.

2.5 Use of neural network methods

2.5.1 Definition

Neural networks are the systems that have emerged due to attempts to simulate the biological processes of cognition. These systems comprise linked neurons that work as functions or simple processing elements. The neurons create a complex system that has a graphical form and can vary significantly in complexity. This means that the networks contain different types of architectures and neurons and have different degrees of complexity. These networks’ particular features depend on the dispositional style, the number of neurons, the presence and design of feedback functions, and specific aims. For example, networks can be designed for specific purposes such as supervised learning, unsupervised learning, optimization procedures, or modeling systems dynamic over time. Neural networks are very flexible and can be used for various applications and problems.

2.5.2 Functioning

Neurons in a neural network perform several functions:

  1. Receive signals from neighboring neurons.

  2. Process these signals using activation functions and weightings.

  3. Generate a nerve impulse or output signal based on the processed signals.

  4. Transmit the output signal to other neighbouring neurons.

2.5.3 Multilayer perceptron

In this scenario, we will consider basic network architecture composed of supervised learning modules known as multilayer perceptrons (MLPs), which are described as static networks lacking any form of feedback loops. This explanation does not incorporate dynamic systems with feedback loops or networks such as Kohonen maps or self-organizing maps, for these are unsupervised classification schemes.

Multilayer perceptrons (MLPs) consist of several neuron layers organized, so that each layer’s output serves as the next layer’s input. The signals to be processed enter via the input layer, which is subdivided into inputs corresponding to neurons.

The output layer produces the system’s response. In this context, the input layer is not considered a separate layer, since it does not introduce any modifications. One or more hidden layers carry out the crucial task of information processing and transfer within the network. Neurons in a hidden layer receive inputs from every neuron in the preceding layer and transmit outputs to each neuron in the subsequent layer. This interconnected structure enables the flow of information throughout the MLP. Figure 1 illustrates an elementary multilayer perceptron consisting of a hidden layer and an output layer.

Figure 1: An example of an elementary multilayer perceptron with a hidden and output layer is.
Figure 1:

An example of an elementary multilayer perceptron with a hidden and output layer is.

2.6 Experimental study

This study aims to calculate and predict thread consumption for technical textile products used in the automotive industry. Figure 2 depicts the analysed component, highlighting the need for precise estimation of sewing thread usage to enhance material efficiency and reduce production costs. By examining its design, stitching patterns, and material properties, this study seeks to create a reliable prediction model for thread consumption, ultimately improving resource management and manufacturing accuracy.

Figure 2: Gear shift shifter cover.
Figure 2:

Gear shift shifter cover.

Table 1 shows the assembly range of the article shown in Figure 2, a leatherette car gearbox cover. The following table illustrates the recording of the various parameters and the consumption of each operation.

Table 1:

Assembly range for “gear shift shifter cover”.

Operation Type of stitch Stitch count/cm Thickness (mm) Linear density (Tex) Length (cm) Layer number Consumption (cm)
Assemble part 1 with part 2 301 2.5 1.08 30/15 32.5 2 239.4
Assemble part 2 with part 3 301 2.5 1.08 30/15 32.5 2 239.5
Assemble part 3 with part 4 with reinforcement 301 2.5 1.08 30/15 32.5 3 269.5
Fix slit 301 2.5 1.08 30/15 2.5 2 35
Topstitch side 1 with lining 301 2.5 1.08 30/15 32.5 3 319.9
Topstitch side 2 with lining 301 2.5 1.08 30/15 32.5 3 319.9
Topstitch side 3 301 2.5 1.08 30/15 32.5 3 318
Close side + topstitch last side 301 2.5 1.08 30/15 32.5 2 274.4
Tayrib assembly (plastic) 301 2.5 1.08 30/15 24.5 2 84.6
Elastic assembly 301 2.5 1.08 30/15 55 2 138.3

The investigated sample, a leatherette car gearbox cover, is typically made from synthetic leather. This material is designed to mimic the appearance and texture of genuine leather, while being more durable and easier to maintain.In terms of the type of fabric used, leatherette is often produced using a substrate like polyester or cotton, which is then coated with a layer of PVC (polyvinyl chloride) or polyurethane (PU). This combination offers a soft feel similar to leather, while providing resistance to wear and tear, making it suitable for automotive applications. Below in Table 2 are the main properties of the sewing threads used in our study.

Table 2:

The main properties of the sewing threads.

Property of sewing thread Value
Linear density 34 tex
Composition 100 % cotton
Breaking strength 8.83 N
Breaking elongation 5.85 %
Rigidity 383.62 N/m
Twists 572 R/m
Tenacity 25.97C N/tex

The Juki DNU 1541 industrial flat bed sewing machine used in our study is a flat bed stitcher with a double capacity hook and a lockstitch mechanism. It is equipped with a remote motor and can reach a maximum stitch speed of 2,500 stitches per minute, which makes it very suitable for industrial applications. The machine has adjustable stitch length and width, with a maximum stitch length of 9 mm. Its rectangular feed mechanism, known as “Box Feed”, improves fabric handling, while the triple feed system ensures that materials are fed consistently. Also, the machine has a table size of 105 × 55 cm, which is quite spacious for different sewing operations.

3 Results and discussions

3.1 Presentation of experimental data for the case study

As shown in Table 3, there is an extensive statistical evaluation of the given parameters alongside the actual useful sewing threads, including various statistical components such as mean, standard deviation, skewness for value and many more to show the users how many parameters can fluctuate. With such raw and powerful information at users’ disposal, it is often easy to analyze the rationale behind the performance and efficiency of sewing threads in real life. This makes it easier to tailor intelligent, automated decision-making in production.

Table 3:

Statistical analysis of input parameters and actual consumption of sewing threads.

Statistical parameter Linear density (Tex) Stitch count/cm Thickness (mm) Number of layers Target consumption (cm)
Mean 46.00 3.00 0.77 2.50 24.58
Standard error 1.69 0.14 0.04 0.08 0.48
Confidence interval 46.00 ± 1.69 3.00 ± 0.14 0.77 ± 0.04 2.50 ± 0.08 24.58 ± 0.48
Median 46.00 3.00 0.76 2.50 24.00
Mode 36.00 2.00 0.76 2.00 24.00
Standard deviation 10.14 0.83 0.25 0.51 2.88
Sample variance 102.86 0.69 0.06 0.26 8.29
Kurtosis (coefficient of flatness) (2.12) (1.54) (1.54) (2.12) 2.17
Skewness coefficient 0.00 (0.00) 0.09 (0.00) 1.60
Range 20.00 2.00 0.60 1.00 11.80
Minimum 36.00 2.00 0.48 2.00 21.20
Maximum 56.00 4.00 1.08 3.00 33.00
Sample size 36.00 36.00 36.00 36.00 36.00
Confidence level (95.0 %) 3.43 0.28 0.08 0.17 0.97

This statistical analysis sheds further light on the construction of the leatherette car gearbox cover. The averages are 46.00 for linear density, 3.00 for stitch count, 0.77 for thickness, 2.50 for Layer number, and 24.58 for target consumption. The mean estimates have a standard error, which signifies a median value closely tied to the mean, suggesting that the middle part of the data is balanced towards the two ends. Values that occur most commonly and are noticed from mode are noted, including those for the linear density and stitch count. Also, there are standard deviation and variance, which illustrate the difference. The linear density has the most incredible spread, meaning there is more deviation from the mean. In this case, the sample size remained constant at 36 and offered reliable analysis. The minimum and maximum values mark the spread of the data. Finally, the 95.0 % mark offers confidence in their reasoning behind such means. These statistics give precision and thus signify how well and poorly the assembly process is done.

3.2 Thread consumption prediction through multiple linear regression

3.2.1 Fixed factors and their levels in thread consumption prediction analysis

Concerning automotive technical products, this section will focus on the non-flexible elements that impact the consumption of threads. Knowing these fixed factors and their corresponding values is critical for thread consumption forecasting. The specific fixed factors incorporated in this analysis and their values are discussed to analyse the impact on thread consumption. Analyzing these fixed factors is intended to articulate better the connection between design parameters and the consumption of threads in automotive engineering.

Indeed, as shown in Table 4, four input parameters are involved in the assembly process of a leatherette car gearbox cover: Linear density, Stitch Count, Thickness, and Number of layers. All factors are classified as fixed, meaning they do not changethroughout the experiment. The Linear density has two levels, 36 and 56. Both the stitch sount and thickness input parameters have three levels; the takes the values of 2, 3, and 4 and the thickness takes the values of 0.48, 0.76, and 1.08. The Layer number factor has two levels of 2 and 3. Totally, there were a total of 36 measurements conducted for the regression analysis. It is possible to evaluate these factors, with their levels and values, showing how the assembly is affected, thusappreciating its quality and performance.

Table 4:

Fixed Factors and their levels.

Input parameter Type Levels Values
Linear density Fixed 2 36, 56
Stitch count Fixed 3 2, 3, 4
Thickness Fixed 3 0.48, 0.76, 1.08
Number of layers Fixed 2 2, 3

3.2.2 Analysis of Variance (ANOVA) in predicting thread consumption

In this part, we will use Analysis of Variance (ANOVA) to determine the causes of thread consumption inautomotive technical products. ANOVA is a sophisticated statistical technique that enables us to evaluate the impactof several factors on thread consumption and determine if the differences seen are significant. Implementing ANOVAwill help us understand which factors are the most impactful in causing differences in thread consumption andimprove our understanding of design variables and material utilization. Such analysis will improve efficiency inproduction technology and resources in the automotive industry.

The variance analysis (ANOVA) and regression analysis presented in Table 5 reveal several key insights. The source of variation is divided into regression – representing the effects of independent variables (linear density, stitch count, thickness, layer number) on the dependent variable (actual consumption)-and error, which captures the unexplained variation. The regression model has 4 degrees of freedom, while the error accounts for 31 degrees of freedom, leading to 35 degrees of freedom. The adjusted sum of squares for regression is 221.849, indicating the total variance explained by the model, whereas the error sum of squares is 68.381, reflecting the variance not explained. The mean square is calculated by dividing the sum of squares by the respective degrees of freedom, and the overall F-value of 25.14 is significant (p-value = 0.000), suggesting that at least one independent variable significantly affects the dependent variable. Specifically, the factor “Linear density” shows an F-value of 22.850 and a p-value of 0.000, indicating a significant effect on actual consumption. Similarly, “Stitch count” has an F-value of 39.800 and a p-value of 0.000, while “thickness” has an F-value of 36.710 and a p-value of 0.000, demonstrating significant impacts. In contrast, the “Layer number” factor presents an F-value of 1.21 and a p-value of 0.280, indicating that it does not significantly affect actual consumption.

Table 5:

Analysis of Variance (Without interactions between factors).

Source DF Adj SS Adj MS F-value P-value
Regression 4 221.849 55.462 25.140 0.000
Linear density 1 50.410 50.410 22.850 0.000
Stitch count 1 87.784 87.784 39.800 0.000
Thickness 1 80.987 80.987 36.710 0.000
Layer number 1 2.668 2.668 1.210 0.280
Error 31 68.381 2.206 – –
Total 35 290.230 – – –

The model summary shown in Table 6 provides important insights into the regression analysis. The standard error (S) is 1.48521, indicating the average distance of observed values from the regression line. The R-squared value of 76.44 % suggests that the independent variables can explain approximately 76.44 % of the variability in actual consumption. The adjusted R-squared of 73.40 % accounts for the number of predictors in the model, while the predicted R-squared of 67.87 % reflects the model’s predictive capability. In terms of coefficients, the constant term is 7.31, representing the intercept of the regression equation. For the independent variables, the coefficient for “Linear density” is 0.1183, indicating that with each unit increase in Linear density, real consumption increases by 0.1183 units, assuming all other factors remain constant. The coefficient for “Stitch count” is 1.912, signifying a substantial increase in real consumption with each unit increase. The coefficient for “Thickness” is 6.12, demonstrating a strong positive effect on real consumption. In contrast, the coefficient for “Layer number” is 0.544, suggesting a minor increase in real consumption, which is not statistically significant (p = 0.280).

Table 6:

Coefficients summary for regression analysis without interaction.

Term Coefficient (Coef) Standard error (SE Coef) T-value P-value VIF
Constant 7.310 2.080 3.510 0.001 –
Linear density 0.118 0.025 4.780 0.000 1
Stitch count 1.912 0.303 6.310 0.000 1
Thickness 6.120 1.010 6.060 0.000 1
Layer number 0.544 0.495 1.100 0.280 1

Based on the regression analysis, real consumption can be predicted using a multiple linear regression the intercept being 7.31 and the coefficients being: for the independent variable “Thickness”, the coefficient is 6.12; for “Stitch count”, 1.912; for “Linear density”, 0.1183; while for “Layer number”, it is 0.544 these all denoting positive effects on real consumption Then again, the real consumption determinants dominate consumption and investment dependencies, thus facilitating the economy’s expansion. The Layer number coefficient, however, did not reflect a statistical significance change (p = 0.280), marking it together with the constant as the non-influential variables.

3.2.3 Understanding main effects on real consumption

Main effects plots serve as a valuable tool for visually summarizing the impact of each factor on the dependent variable. By examining these plots, we can gain insights into the key drivers of real consumption. This analysis enables a clearer understanding of how various factors interact and influence outcomes, ultimately guiding informed decision-making based on the findings.

In main effects plots as presented in Figure 3, evaluating the impact of each factor on the dependent variable, while controlling for other variables is a significant consideration. Correspondingly, the stitch count plot captures average real consumption for certain stitch count levels (2, 3, 4). The stitch count increments, represented by the slope, correspondto an increased consumption average at the consumption stitch counts. The thickness plot attempts to portray therelationship between the real consumption and thickness levels (0.48, 0.76, 1.08), and it is seen that thicker materialspositively increase the level of consumption.In contrast, the Layer number plot compares two levels (2 vs. 3) and shows minimal differences in average consumption, supporting the ANOVA finding that Layer number does not significantly affect consumption. General considerations when interpreting these plots include the slope’s steepness-indicating the strength of the effect-the statistical significance of differences between levels as determined by ANOVA p-values, interaction effects indicated by non-parallel lines in interaction plots, and visual representations that often include error bars to illustrate variability.

Figure 3: Main effects plot for target consumption.
Figure 3:

Main effects plot for target consumption.

3.2.4 Analyzing the contour plot: target thread consumption insights

Interpreting a contour plot of target consumption about Layer number and thickness as shown in Figure 4 involves examining the interaction between these two factors and their impact on target consumption. The x-axis typically represents thickness, while the y-axis represents layer number, with the contour lines indicating target consumption levels, connecting stitch counts of equal consumption. Close contours suggest a steep gradient, meaning small changes in either factor can significantly affect consumption, whereas wider spacing indicates a lesser impact. The shape of the contours further reveals the nature of the relationship: linear contours imply a consistent relationship. In contrast, curved or irregular contours indicate more complex interactions, suggesting that the effect of Thickness on consumption may vary with layer number levels. Contours at higher levels are regions where higher target consumption values are found, and these regions depict optimal combinations of Thickness and Layer number input parameters. In contrast, less desirable combinations correspond with low consumption areas. Most markedly, some changes in contour shape or steepness suggest factors that change the slope, with other factors modulating the effect on consumption. Using the contour plot, considerable value can be analyzed in terms of determining where slice and Thickness settings that meet target consumption levels can enhance decision-making on the process or product design to improve operational effectiveness and efficiency.

Figure 4: Contour plots of target consumption versus input parameters. (a)Target consumption versus layer number and thickness (b) target consumption versus linear density and stitch count.
Figure 4:

Contour plots of target consumption versus input parameters. (a)Target consumption versus layer number and thickness (b) target consumption versus linear density and stitch count.

3.2.5 Key features of a residual diagram

As crucial as the rest of the regression steps, interpreting the residual diagram, or residual plot, as shown in Figure 5 is equally important when validating a regression model. Typically, the x-axis depicts the predicted results (which are the fitted values from the model), and the y-axis depicts the residuals, which is the difference between the observed and state values. Often, a horizontal line at zero is used to visualize the centring of the residuals, although the best situation is where the residuals are randomly distributed around this horizontal line. Randomness in the residuals suggests that the model has successfully captured the relationship between independent and dependent variables with the linearity and Homoscedasticity (an assumption that there is constant variance) that it requires. Conversely, any curves or trends which are discernible patterns indicate that the model may need to be less straightforward. Without taking other predictors, a non-linear approach may be necessary. Further, any systematic changes in the spread of the residuals indicate funnelling or disruption of heteroscedasticity. Regression assumptions are violated in this case. Outliers, or observations that are distant from the cluster of residuals and can drastically change the model’s results as opposed to other classes of the portion that lie within the zone in which the average deviates, should also be looked into closely. Although the residual normality may not be evident in the plot, a histogram or Q-Q plot can provide additional information; in an ideal world, they could support inference by displaying a normal distribution. On the other side, a reasonable residual plot tells us that the regression model is probably correct, but the presence of systematic deviations from the horizontal line suggests that some corrective actions should be made; for example, the model can be enriched by adding more factors or different modeling approaches can be searched. Understanding outliersis important, so that these stitch counts do not adversely affect the analysis, and the residual plot offers excellent value in understanding how well the regression model works and what can be changed to better it.

Figure 5: Residual diagram.
Figure 5:

Residual diagram.

3.2.6 Results and model summary

The Table 7 analyzes some of the most relevant metrics of a regression model deemed to assess its quality. S represents the standard error set at 1.485, which portrays the deviation of the predicted values from the data stitch counts graphed on the regression line, indicating a relatively poor fit.The R2 value of 76.44 % indicates that approximately 76.44 % of the variance in the dependent variable is explained by the independent variables, demonstrating a relatively strong relationship. The adjusted R2 of 73.4 % accounts for the number of predictors, confirming that a significant portion of variance is still explained despite this adjustment. The Root Mean Square Error (RMSE) of 1.378 measures the average magnitude of prediction errors, while the mean absolute error (MAE) of 1.075 indicates the average absolute deviation of predictions from actual values. These statistics suggest that the model performs reasonably well, though there may still be opportunities for improving prediction accuracy.

Table 7:

Model summary statistics (without interactions between factors).

S R2 (%) R2 Adj (%) RMSE MAE
1.49 76.44 73.4 1.38 1.08

Figure 6 illustrates the comparison between target and predicted consumption, showing a linear relationship characterized by the equation ( y = 0.76 x+5.79) and a coefficient of determination (R2) of 0.76. This indicates that the target values can explain approximately 76 % of the variance in the predicted values. However, this R2 value is deemed unsatisfactory, leading to the consideration of developing a linear regression model that incorporates interactions between the factors.

Figure 6: Comparison of target and predicted consumption (without interactions between factors).
Figure 6:

Comparison of target and predicted consumption (without interactions between factors).

3.3 Incorporating factor interactions in multiple linear regression models

3.3.1 Analysis of variance

The ANOVA as shown in Table 8 summarizes the results of an Analysis of Variance for a multiple linear regression model that includes factor interactions. It outlines various sources of variation, including the overall regression model, individual factors (linear density, stitch count, thickness, layer number), and their interactions. The degrees of freedom (DF) indicate the number of independent values that can vary, with the regression having 12 DF and the error term 23. The Adjusted Sum of Squares (Adj SS) shows how much variation each source explains, with the regression model accounting for a total of 263.614 units of variation. At the same time, specific contributions include a stitch count of 12.907 and a Thickness of 5.182. The Adjusted Mean Square (Adj MS) provides an average measure of this variation, and the F-Value tests the null hypothesis that the means are equal; higher values suggest significant relationships, particularly for stitch count (F = 11.15) and the interaction linear density × stitch count (F = 13.41). The P-value indicates the probability of observing the data under the null hypothesis, with values below 0.05 signifying significant effects; for instance, stitch sount has a P-value of 0.003, and Linear density × Stitch count has a P-value of 0.001, indicating strong significance. In contrast, the Layer number shows no significant contribution with a P-value of 0.956. Overall, the regression model explains a substantial variation in the response variable, aiding in identifying important predictors for future research and model refinement.

Table 8:

Analysis of Variance (including interactions between factors).

Source DF Adj SS Adj MS F-value P-value
Regression 12 263.614 21.968 18.980 0
Linear density 1 4.068 4.068 3.520 0.074
Stitch count 1 12.907 12.907 11.150 0.003
Thickness 1 5.182 5.182 4.480 0.045
Layer number 1 0.004 0.004 0.000 0.956
Linear density × stitch count 1 15.520 15.520 13.410 0.001
Linear density × thickness 1 8.707 8.707 7.520 0.012
Linear density × layer number 1 0.250 0.250 0.220 0.646
Stitch count × thickness 1 9.637 9.637 8.330 0.008
Stitch count × stitch count 1 6.301 6.301 5.450 0.029
Stitch count × layer number 1 0.094 0.094 0.080 0.778
Thickness × thickness 1 0.674 0.674 0.580 0.453
Thickness × layer number 1 0.581 0.581 0.500 0.486
Error 23 26.616 1.157 – –
Total 35 290.230 – – –

Table 9 presents the coefficients summary for the regression analysis, including interaction terms. The “Term” column lists the predictors, including the constant, individual factors (linear density, stitch count, thickness, layer number), and their interactions. The “Coef” column shows the estimated coefficients for each term, indicating the average change in the response variable for a one-unit change in the predictor. For instance, the coefficient for stitch count is −9.42, suggesting that an increase in stitch count is associated with a decrease in the response variable. The “SE Coef” column presents the standard error of each coefficient, with the constant having a standard error of 8.6, reflecting the uncertainty in the estimate.

Table 9:

Coefficients summary for regression (including interactions between factors).

Term Coef SE Coef T-value P-value VIF
Constant 41.760 8.600 4.860 0.000 –
Linear density −0.236 0.126 −1.880 0.074 49.450
Stitch count −9.420 2.820 −3.340 0.003 165.110
Thickness −18.540 8.760 −2.120 0.045 143.450
Layer number 0.130 2.420 0.060 0.956 45.610
Linear density × stitch count 0.080 0.022 3.660 0.001 35.660
Linear density × thickness 0.201 0.073 2.740 0.012 32.110
Linear density × layer number −0.017 0.036 −0.460 0.646 47.160
Stitch count × thickness 2.585 0.896 2.890 0.008 24.450
Stitch count × stitch count 0.888 0.380 2.330 0.029 109.000
Stitch count × layer number 0.125 0.439 0.280 0.778 39.500
Thickness × thickness 3.240 4.250 0.760 0.453 83.790
Thickness × layer number 1.040 1.460 0.710 0.486 35.950

The “T-Value” column indicates the coefficient’s ratio to its standard error, which is used to assess the significance of each term. Notably, Linear density has a T-Value of −1.88 with a P-Value of 0.074, indicating marginal significance. At the same time, Stitch Count shows a T-value of −3.34 and a P-value of 0.003, suggesting a significant negative impact on the response variable. Thickness also shows significance with a T-value of −2.12 and a P-value of 0.045.

The interaction term linear density × stitch count has a coefficient of 0.0804 and a P-value of 0.001, indicating a significant positive interaction effect. Conversely, the interaction Linear density × Layer number is insignificant, with a P-value of 0.646. The “VIF” (Variance Inflation Factor) column assesses multicollinearity, where values above 10 may indicate potential issues; for example, stitch count has a VIF of 165.11, suggesting high multicollinearity. This table provides insights into the significance and impact of various factors and their interactions on the response variable, highlighting both significant and non-significant predictors.

3.3.2 Interactions plot for target consumption (including interactions between factors)

The interaction plot as shown in Figure 7 illustrates the effects of various factors-linear density, stitch count, thickness, and layer number-on target consumption. It reveals that consumption rises with increasing values of stitch count and Thickness, with Linear density 56 associated with higher consumption levels compared to Linear density 36. Notably, the interaction between stitch count and thickness is pronounced, suggesting that thicker materials combined with higher stitch count values significantly increase consumption. In contrast, the influence of the Layer number appears minimal, remaining relatively consistent across different scenarios. Overall, the findings emphasize that stitch count and Thickness are the key determinants of target consumption.

Figure 7: Interaction plot for target consumption.
Figure 7:

Interaction plot for target consumption.

3.3.3 Results and model summary

Table 10 and Figure 8 present the model summary statistics, which include key metrics for evaluating the performance of the regression model that incorporates interactions between factors. The standard deviation (S) of 0.8721 indicates the average distance between observed and predicted values, suggesting reasonable accuracy. The correlation coefficient (R) of 95.30 % reflects a strong linear relationship between predicted and actual consumption values. The R2 value of 90.83 % indicates that the model can explain approximately 90.83 % of the variability in consumption, while the adjusted R2 of 89.97 % accounts for the number of predictors, reinforcing the model’s validity. Additionally, the root mean square error (RMSE) of 0.8601 and the mean absolute error (MAE) of 0.7545 suggest acceptable levels of prediction error. However, there remains room for improvement in the consumption prediction model. We will explore artificial intelligence methods to enhance its accuracy, mainly focusing on testing neural networks in the following section.

Table 10:

Model summary statistics (including interactions between factors).

S R R2 R2 (adj) RMSE MAE
0.87 95.30 90.83 % 89.97 % 0.86 0.75
Figure 8: Comparison of target and predicted consumption (including interactions between factors).
Figure 8:

Comparison of target and predicted consumption (including interactions between factors).

3.4 Sewing thread consumption prediction using neural networks

3.4.1 Presentation of the model used

Figure 9 describes a neural network model with four input and one output neuron, one hidden layer with 10 neurons, a hidden layer’s activation function of logistic sigmoid ‘logsig’, and an output layer’s ‘purelin’ linear activation function. To train the model, the Bayesian regularization training function is used to bias the learning in favor of generalization and to prevent overfitting. A maximum of 1,000 epochs is set to give the network enough chances to converge to a solution. At the same time, an error performance target of le-5 ensures sufficient accuracy in the predictions the model gives as output.

Figure 9: The neural network configuration.
Figure 9:

The neural network configuration.

3.4.2 Evaluation of test and training phase

The graph “Best Training Performance” as presented in Figure 10 captures the entire view of the deep learning model’s training behavior. The mean square error (MSE), a critical measure of how well the model functions in the training and testing stages, is represented at the Y-axis. At the X-axis, there are the iterations or epochs, which show the model’s performance concerning time. Significantly, the training performance curve is improving with each iteration, which indicates that the model is learning from the training data. In epoch 135, the training performance peaked at 0.012519, implying that a remarkable error minimisation progress has been achieved. This graph informs of the results relative to the neural network training, offering information on its learning efficiency and monitoring. It also emphasizes the need to monitor testing performance, so that the model is not only learned, but can also generalize effectively. In summary, while the model continuously improves performance through training, it increases the risk of overfitting, which calls for supervision. In the evaluation step of the neural network performance, a learning process usually involves training the network and models during a specific time period using a training data set. A different test data set is reserved for validating the results, which permits the juxtaposition of the actual values against the predicted values.

Figure 10: Best training performance.
Figure 10:

Best training performance.

3.4.3 Results of the prediction using neural network

In the evaluation step of the neural network performance, a learning process usually involves training the network and models during a specific time period using a training data set containing 25 samples. A different test data set containing 11 samples is reserved for validating the results, which permits the juxtaposition of the actual values against the predicted values as presented in Figure 11.

Figure 11: Analysis of target versus predicted consumption in training and test stages.
Figure 11:

Analysis of target versus predicted consumption in training and test stages.

As with any advanced algorithm, verifying its predictions and defining success across metrics is essential, and in the case of the neural net validation results, it was successful at accuracy across all metrics. According to Table 11, with a Test RMSE of 0.60468, the model shows a reasonable average deviation within an acceptable range, as evidenced by the estimated deviation of 0.43264 in Test MAE. Likewise, the model’s ability to capture relationships in the data is shown through the acceptable strong positive correlation of 0.97182 in the Test Correlation Coefficient (R) and the predicted values, which align perfectly with the real values. This is further demonstrated through the 0.94442 Adjusted R2(R2 adj) and shows even more trustable results, because it is tested, suggesting that the model can explain approximately 94.44 % of the variance in the test data. In addition, the model proved strong explanatory power with the calculations fromnumber of explanation variables accepted R-squared which was 0.90737. Our results show better performance than other thread consumption prediction studies. Our results show better performance than previous studies even though the textile support in our study is different because we focus on technical textiles in the automotive sector. All these observations indicate that the neural network could learn at an appropriate level for elasticity and thus can be considered an accurateforecasting method. However, some degree of follow-up will be required.

Table 11:

Test data performance metrics.

Metric Value
RMSE 0.60
MAE 0.43
Correlation coefficient (R) 0.97
(R2) 0.94
R2 Adjusted 0.91

3.5 Validation of results

We need to validate the results obtained previously by testing our neural model which has shown excellent results with other products in the automotive sector. For example, we will take the automotive straps, illustrated in Figure 12.

Figure 12: Automotive straps.
Figure 12:

Automotive straps.

Table 12 presents the values of the input parameters used in our analysis, including three levels of linear density (36, 44, 56), three options for the number of stitches (2, 3, 4), as well as three thickness values (0.55, 0.85, 1.20 mm) and two levels for the number of layers (2 and 3). These parameters are essential to assess their respective impact on the performance of the product studied in the automotive sector, allowing an in-depth analysis of their interactions and their effects on the model results.

Table 12:

The values of the input parameters.

Input parameter Levels Values
Linear density 3 36, 44, 56
Stitch count 3 2, 3, 4
Thickness 3 0.55, 0.85, 1.20
Number of layers 2 2, 3

We maintained the same neural network model throughout our study by using parameters. The database contains 54 experimental tests where 38 samples were used for model training and 16 samples were set aside for testing to assess model performance. The neural network predictions match experimental values as shown in Figure 13 which demonstrates the model’s prediction effectiveness. The statistical results from this study are presented in Table 12 which provides exact evaluation metrics for both accuracy and reliability of the model.

Figure 13: Comparison of target and predicted consumption.
Figure 13:

Comparison of target and predicted consumption.

The performance metrics of our model for test data are shown in Table 13 and Figure 13. The RMSE (Root Mean Square Error) is 1.1617, which shows the average difference between the predicted and observed values. The MAE (Mean Absolute Error) is 1.0075, which means that our predictions are, on average, off by about 1 unit from the actual values. The correlation coefficient (R) is 0.94413, which shows that there is a strong relationship between the predicted and observed values. The R2 (coefficient of determination) is 0.89138, which means that about 89 % of the variance in the data can be explained by the model. The adjusted R2 of 0.85189, which takes into account the number of model parameters, still suggests that the performance is good even after adjustment. These results indicate that the model works well and is reliable for the tested data.

Table 13:

Test data performance metrics.

Metric Value
RMSE 1.16
MAE 1.01
Correlation coefficient (R) 0.94
(R2) 0.89
R2 Adjusted 0.85

4 Conclusions

The neural network model developed for this research demonstrated reliable performance in predicting automotive sewing thread usage. The obtained results demonstrate strong correlation coefficients, which indicate an excellent match between predicted and actual values. Our model demonstrates performance capabilities that make it useful for automotive industry process optimization through precise and meaningful forecasting.

In summary, this research analysed the effects of specific approach parameters such as seam length, stitch density, fabric thickness, and stitch type on thread consumption, while sewing. A neural network was used to estimate the thread consumption. The system can accurately predict thread consumption with sufficient training on relevant features and patterns. This is useful information for an organization that needs proper resource planning and threading optimization. In any case, applying a neural network in this domain causes an improvement in the productivity and accuracy of the estimated value of the thread consumption. In general, this study broadens the scope of understanding regarding the reasons for the consumption of sewing thread. It establishes reliable approaches for planning thread consumption in different sewing processes. The results can assist in minimizing the consumption of thread, augmenting cost-effectiveness, and maximizing productivity in sewing activities.

Corresponding author: Thouraya Hamdi, Department of Fashion and Textile Design, College of Arts and Design, Princess Nourah bint Abdulrahman University, P.O. Box 84428, Riyadh, 11671, Saudi Arabia, E-mail:

Acknowledgements

Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2025R246), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

  1. Funding information: Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2025R246), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

  2. Author contributions: The authors took full responsibility for the entire content of this manuscript and gave their consent for journal submission while reviewing all results before approving the final manuscript version. Thouraya Hamdi, Faouzi Khedher, Mohamed Jmali and Boubaker Jaouachi contributed to the writing, review, investigation, analysis, and validation of the manuscript.

  3. Conflict of interest: Authors state no conflict of interest.

  4. Ethical approval: The conducted research is not related to either human or animals use.

  5. Data Availability Statement: The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

References

[1] K. Niinimäki and L. Hassi, “Emerging design strategies in sustainable production and consumption of textiles and clothing,” J. Clean. Prod., vol. 19, no. 16, pp. 1876–1883, 2011.10.1016/j.jclepro.2011.04.020Search in Google Scholar

[2] T. Karthik, P. Ganesan, and D. Gopalakrishnan, Apparel Manufacturing Technology, CRC Press, 2016.10.1201/9781315367507Search in Google Scholar

[3] Y. Xu, S. Thomassey, and X. Zeng, “AI for apparel manufacturing in big data era: a focus on cutting and sewing,” Artif. Intell. Fash. Ind. Big Data Era, pp. 125–151, 2018, https://doi.org/10.1007/978-981-13-0080-6-7.Search in Google Scholar

[4] M. Jaouadi, S. Msahli, A. Babay, and B. Zitouni, “Analysis of the modeling methodologies for predicting the sewing thread consumption,” Int. J. Clot. Sci. Technol., vol. 18, no. 1, pp. 7–18, 2006, https://doi.org/10.1108/09556220610637477.Search in Google Scholar

[5] J. Bullon, M. A. González Arrieta, A. Hernández Encinas, and M. A. Queiruga Dios, “Manufacturing processes in the textile industry,” Expert Syst. Fabr. Prod., vol. 6, no. 1, pp. 41–50, 2017.10.14201/ADCAIJ2017614150Search in Google Scholar

[6] E. Gökalp, M. O. Gökalp, and P. E. Eren, “Industry 4.0 revolution in clothing and apparel factories: apparel 4.0,” Industry, vol. 4, pp. 169–183, 2018.Search in Google Scholar

[7] M. V. Chavhan, M. R. Naidu, and H. Jamakhandi, “Artificial neural network and regression models for prediction of sewing thread consumption for multilayered fabric assembly at lockstitch 301 seam,” Res. J. Textile Appar., vol. 26, no. 4, pp. 343–358, 2022, https://doi.org/10.1108/rjta-09-2020-0103.Search in Google Scholar

[8] R. P. Abeysooriya and G. L. D. Wickramasinghe, “Regression model to predict thread consumption incorporating thread-tension constraint: study on lockstitch 301 and chain-stitch 401,” Fash. Textil., vol. 1, no. 1, p. 14, 2014, https://doi.org/10.1186/s40691-014-0014-5.Search in Google Scholar

[9] B. Jaouachi and F. Khedher, “Evaluating sewing thread consumption of Jean pants using fuzzy and regression methods,” J. Text. Inst., vol. 104, no. 10, pp. 1065–1070, 2013, https://doi.org/10.1080/00405000.2013.773627.Search in Google Scholar

[10] C. Karmaker, P. K. Halder, and E. Sarker, “A study of time series model for predicting jute yarn demand: case study,” J. Ind. Eng., vol. 2017, no. 1, pp. 1–8, 2017. https://doi.org/10.1155/2017/2061260.Search in Google Scholar

[11] S. Malek, F. Khedher, D. C. Adolphe, and B. Jaouachi, “Sewing thread consumption for chain stitches of class 400 using geometrical and multi-linear regression models,” Autex Res. J., vol. 21, no. 1, pp. 52–62, 2021, https://doi.org/10.2478/aut-2019-0051.Search in Google Scholar

[12] A. Rehman 2021- ur Rehman, A. Rasheed, Z. Javed, M. S. Naeem, B. Ramzan, and M. Karahan, “Geometrical model to determine sewing thread consumption for stitch class 406,” Fibres Textil. East. Eur., vol. 6, no. 150, 2021.10.5604/01.3001.0015.2726Search in Google Scholar

[13] B. Jaouachi, F. Khedher, and D. Adolphe, “Compared basic stitch’s consumptions using image analysis, geometrical modelling and statistical techniques,” J. Textil. Inst., vol. 110, no. 9, pp. 1280–1292, 2019, https://doi.org/10.1080/00405000.2018.1559016.Search in Google Scholar

[14] B. Jaouachi and F. Khedher, “Evaluation of sewed thread consumption of Jean trousers using neural network and regression methods,” Fibres Textil. East. Eur., vol. 3, no. 111, pp. 91–96, 2015, https://doi.org/10.5604/12303666.1152518.Search in Google Scholar

[15] M. Sarah, J. Boubaker, and A. C. Dominique, “Determination of sewing thread consumption for 602, 605, and 607 cover stitches using geometrical and multi-linear regression models,” Autex Res. J., vol. 22, no. 4, pp. 497–508, 2020, https://doi.org/10.2478/aut-2020-0044.Search in Google Scholar

[16] R. Chauhan and S. Ghosh, “Effect of different parameters on stitch shape and thread consumption,” Ind. J. Fibre Textil. Res. (IJFTR), vol. 45, no. 4, pp. 411–418, 2021.10.56042/ijftr.v45i4.28573Search in Google Scholar

[17] M. V. Chavhan, M. R. Naidu, and H. Jamakhandi, “Artificial neural network and regression models for prediction of sewing thread consumption for multilayered fabric assembly at lockstitch 301 seam,” Res. J. Textil. Appar., vol. 26, no. 4, pp. 343–358, 2022, https://doi.org/10.1108/rjta-09-2020-0103.Search in Google Scholar

[18] T. M. A. Sarhan 2013- Sarhan, “Interaction between sewing thread size and stitch density and its effects on the seam quality of wool fabrics,” J. Appl. Sci. Res., vol. 9, no. 8, pp. 4548–4557, 2013.Search in Google Scholar

[19] R. Chauhan and S. Ghosh, “Geometric model of lockstitch seam and prediction of thread consumption,” J. Textil. Inst., vol. 113, no. 2, pp. 314–323, 2022, https://doi.org/10.1080/00405000.2021.1872837.Search in Google Scholar

[20] Z. Javed et al.., “Modeling the consumption of sewing thread for stitch class 301 through image analysis by using fourier series,” Text. Res. J., vol. 92, nos. 15–16, pp. 2792–2799, 2022, https://doi.org/10.1177/00405175211017400.Search in Google Scholar

[21] S. I. Malami, M. M. Jibril, I. A. Mahmoud, U. J. Muhammad, and J. Z. Yau, “Design of machine learning model for predicting the compressive strength of fabric fiber-reinforced Portland cement,” Techno-Comput. J., vol. 1, no. 1, pp. 14–26, 2025, https://doi.org/10.71170/tecoj.2025.1.1.pp14-26.Search in Google Scholar

[22] H. Dau Sy, P. D. Thi, H. V. Gia, and K. L. Nguyen An, “Automated fabric defect classification in textile manufacturing using advanced optical and deep learning techniques,” Int. J. Adv. Des. Manuf. Technol., vol. 137, pp. 1–15, 2025, https://doi.org/10.1007/s00170-025-15342-z.Search in Google Scholar

[23] B. Shanmughan and B. Kandasubramanian, “Artificial intelligence techniques to predict the behavior of silk fibroin,” in Engineering natural silk: applications and future directions, Singapore, Springer Nature Singapore, 2024, pp. 193–206.10.1007/978-981-97-7901-7_11Search in Google Scholar
Received: 2025-03-21
Accepted: 2025-10-28
Published Online: 2025-12-29

© 2025 the author(s), published by De Gruyter, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

  2. Study and restoration of the costume of the HuoLang (Peddler) in the Ming Dynasty of China
  3. Texture mapping of warp knitted shoe upper based on ARAP parameterization method
  4. Extraction and characterization of natural fibre from Ethiopian Typha latifolia leaf plant
  5. The effect of the difference in female body shapes on clothing fitting
  6. Structure and physical properties of BioPBS melt-blown nonwovens
  7. Optimized model formulation through product mix scheduling for profit maximization in the apparel industry
  8. Fabric pattern recognition using image processing and AHP method
  9. Optimal dimension design of high-temperature superconducting levitation weft insertion guideway
  10. Color analysis and performance optimization of 3D virtual simulation knitted fabrics
  11. Analyzing the effects of Covid-19 pandemic on Turkish women workers in clothing sector
  12. Closed-loop supply chain for recycling of waste clothing: A comparison of two different modes
  13. Personalized design of clothing pattern based on KE and IPSO-BP neural network
  14. 3D modeling of transport properties on the surface of a textronic structure produced using a physical vapor deposition process
  15. Optimization of particle swarm for force uniformity of personalized 3D printed insoles
  16. Development of auxetic shoulder straps for sport backpacks with improved thermal comfort
  17. Image recognition method of cashmere and wool based on SVM-RFE selection with three types of features
  18. Construction and analysis of yarn tension model in the process of electromagnetic weft insertion
  19. Influence of spacer fabric on functionality of laminates
  20. Design and development of a fibrous structure for the potential treatment of spinal cord injury using parametric modelling in Rhinoceros 3D®
  21. The effect of the process conditions and lubricant application on the quality of yarns produced by mechanical recycling of denim-like fabrics
  22. Textile fabrics abrasion resistance – The instrumental method for end point assessment
  23. CFD modeling of heat transfer through composites for protective gloves containing aerogel and Parylene C coatings supported by micro-CT and thermography
  24. Comparative study on the compressive performance of honeycomb structures fabricated by stereo lithography apparatus
  25. Effect of cyclic fastening–unfastening and interruption of current flowing through a snap fastener electrical connector on its resistance
  26. NIRS identification of cashmere and wool fibers based on spare representation and improved AdaBoost algorithm
  27. Biο-based surfactants derived frοm Mesembryanthemum crystallinum and Salsοla vermiculata: Pοtential applicatiοns in textile prοductiοn
  28. Predicted sewing thread consumption using neural network method based on the physical and structural parameters of knitted fabrics
  29. Research on user behavior of traditional Chinese medicine therapeutic smart clothing
  30. Effect of construction parameters on faux fur knitted fabric properties
  31. The use of innovative sewing machines to produce two prototypes of women’s skirts
  32. Numerical simulation of upper garment pieces-body under different ease allowances based on the finite element contact model
  33. The phenomenon of celebrity fashion Businesses and Their impact on mainstream fashion
  34. Marketing traditional textile dyeing in China: A dual-method approach of tie-dye using grounded theory and the Kano model
  35. Contamination of firefighter protective clothing by phthalates
  36. Sustainability and fast fashion: Understanding Turkish generation Z for developing strategy
  37. Digital tax systems and innovation in textile manufacturing
  38. Applying Ant Colony Algorithm for transport optimization in textile industry supply chain
  39. Innovative elastomeric yarns obtained from poly(ether-ester) staple fiber and its blends with other fibers by ring and compact spinning: Fabrication and mechanical properties
  40. Design and 3D simulation of open topping-on structured crochet fabric
  41. The impact of thermal‒moisture comfort and material properties of calf compression sleeves on individuals jogging performance
  42. Calculation and prediction of thread consumption in technical textile products using the neural network method
https://doi.org/10.1515/aut-2025-0062
Keywords for this article
yarn consumption; textile technical products; neural network
Creative Commons
BY 4.0

Articles in the same Issue

  2. Study and restoration of the costume of the HuoLang (Peddler) in the Ming Dynasty of China
  3. Texture mapping of warp knitted shoe upper based on ARAP parameterization method
  4. Extraction and characterization of natural fibre from Ethiopian Typha latifolia leaf plant
  5. The effect of the difference in female body shapes on clothing fitting
  6. Structure and physical properties of BioPBS melt-blown nonwovens
  7. Optimized model formulation through product mix scheduling for profit maximization in the apparel industry
  8. Fabric pattern recognition using image processing and AHP method
  9. Optimal dimension design of high-temperature superconducting levitation weft insertion guideway
  10. Color analysis and performance optimization of 3D virtual simulation knitted fabrics
  11. Analyzing the effects of Covid-19 pandemic on Turkish women workers in clothing sector
  12. Closed-loop supply chain for recycling of waste clothing: A comparison of two different modes
  13. Personalized design of clothing pattern based on KE and IPSO-BP neural network
  14. 3D modeling of transport properties on the surface of a textronic structure produced using a physical vapor deposition process
  15. Optimization of particle swarm for force uniformity of personalized 3D printed insoles
  16. Development of auxetic shoulder straps for sport backpacks with improved thermal comfort
  17. Image recognition method of cashmere and wool based on SVM-RFE selection with three types of features
  18. Construction and analysis of yarn tension model in the process of electromagnetic weft insertion
  19. Influence of spacer fabric on functionality of laminates
  20. Design and development of a fibrous structure for the potential treatment of spinal cord injury using parametric modelling in Rhinoceros 3D®
  21. The effect of the process conditions and lubricant application on the quality of yarns produced by mechanical recycling of denim-like fabrics
  22. Textile fabrics abrasion resistance – The instrumental method for end point assessment
  23. CFD modeling of heat transfer through composites for protective gloves containing aerogel and Parylene C coatings supported by micro-CT and thermography
  24. Comparative study on the compressive performance of honeycomb structures fabricated by stereo lithography apparatus
  25. Effect of cyclic fastening–unfastening and interruption of current flowing through a snap fastener electrical connector on its resistance
  26. NIRS identification of cashmere and wool fibers based on spare representation and improved AdaBoost algorithm
  27. Biο-based surfactants derived frοm Mesembryanthemum crystallinum and Salsοla vermiculata: Pοtential applicatiοns in textile prοductiοn
  28. Predicted sewing thread consumption using neural network method based on the physical and structural parameters of knitted fabrics
  29. Research on user behavior of traditional Chinese medicine therapeutic smart clothing
  30. Effect of construction parameters on faux fur knitted fabric properties
  31. The use of innovative sewing machines to produce two prototypes of women’s skirts
  32. Numerical simulation of upper garment pieces-body under different ease allowances based on the finite element contact model
  33. The phenomenon of celebrity fashion Businesses and Their impact on mainstream fashion
  34. Marketing traditional textile dyeing in China: A dual-method approach of tie-dye using grounded theory and the Kano model
  35. Contamination of firefighter protective clothing by phthalates
  36. Sustainability and fast fashion: Understanding Turkish generation Z for developing strategy
  37. Digital tax systems and innovation in textile manufacturing
  38. Applying Ant Colony Algorithm for transport optimization in textile industry supply chain
  39. Innovative elastomeric yarns obtained from poly(ether-ester) staple fiber and its blends with other fibers by ring and compact spinning: Fabrication and mechanical properties
  40. Design and 3D simulation of open topping-on structured crochet fabric
  41. The impact of thermal‒moisture comfort and material properties of calf compression sleeves on individuals jogging performance
  42. Calculation and prediction of thread consumption in technical textile products using the neural network method
Sign up now to receive a 20% welcome discount
Institutional Access
How does access work?
Have an idea on how to improve our website?
Please write us.
© 2026 De Gruyter Brill
Downloaded on 8.1.2026 from https://www.degruyterbrill.com/document/doi/10.1515/aut-2025-0062/html
Scroll to top button