The Transformative Potential of Industry 4.0 in Textile Manufacturing
The enhancement of fabric quality prediction in the textile manufacturing sector is achieved by leveraging information derived from sensors within the Internet of Things (IoT) and Enterprise Resource Planning (ERP) systems linked to sensors embedded in textile machinery. The integration of Industry 4.0 concepts is instrumental in harnessing IoT sensor data, which, in turn, leads to improvements in productivity and reduced lead times in textile manufacturing processes.
This study addresses the challenge of imbalanced data pertaining to fabric quality within the textile manufacturing industry. It encompasses an evaluation of seven open-source automated machine learning (AutoML) technologies, namely FLAML, AutoViML, EvalML, AutoGluon, H2OAutoML, PyCaret, and TPOT (Tree-based Pipeline Optimization Tool). The most suitable solutions are chosen for certain circumstances by employing an innovative approach that finds a compromise among computational efficiency and forecast accuracy.
Unraveling the Complexities of Fabric Quality Prediction
The results reveal that EvalML emerges as the top-performing AutoML model for a predetermined objective function, particularly excelling in terms of mean absolute error (MAE). On the other hand, even with longer inference periods, AutoGluon performs better than other methods in measures like mean absolute percentage error (MAPE), root mean squared error (RMSE), and r-squared.
Additionally, the study explores the feature importance rankings provided by each AutoML model, shedding light on the attributes that significantly influence predictive outcomes. Notably, sin/cos encoding is found to be particularly effective in characterizing categorical variables with a large number of unique values.
Leveraging Industry 4.0 for Enhanced Productivity and Quality Control
This study includes useful information about the application of AutoML in the textile industry and provides a roadmap for employing Industry 4.0 technologies to enhance fabric quality prediction. The research highlights the importance of striking a balance between predictive accuracy and computational efficiency, emphasizes the significance of feature importance for model interpretability, and lays the groundwork for future investigations in this field.
The Evolving Landscape of Textile Manufacturing
Textile manufacturing firms are under constant pressure in the current competitive market from rising consumer expectations for customized items that exhibit a higher quality while also having a reduced production cost. Following the orders that are received and the designs that have been made, production follows the planning process. If the final product fulfills the requirements of the quality control process, it is delivered. However, the production procedure and the final product must be routinely examined by the quality control process, and any errors must be fixed and new plans must be made.
To meet the objectives of quicker delivery times and better textile quality in this situation, adopting intelligent integration solutions is the best course of action for the entire supply chain. The typical method is to collect data from the ERP system and then transfer the data to the cloud platform for statistical analysis. The information gathered relates to the characteristics of the fiber, the process parameters, the yarn, the requirements for the loom, and the machine feature taken from the firm ERP.
Challenges in Textile Manufacturing Processes
Raw materials go through a number of processes where fibers and yarns are joined, and the combination of these yarns goes through a number of textile processes, eventually generating a fabric as its end product. In numerous stages of this production process, automation technologies might be used. In the loom’s operation, the insertion of weft threads into warps is a fundamental step. This process involves introducing the weft, threading the needle via the fabric’s shedding, and utilizing the reed pulses to move the inserted thread across the fabric that has already been created.
The warp yarns may break as a result of this procedure. The pressure that the yarn is put under during this process will show where the basic material is weak. As a result, while under tension, a yarn with thin spots would usually break, as opposed to other spots on the yarn. The amount of friction between the threads will rise as the process speed rises. Because of this, the procedure will get more tense, which will result in more breaks. In the production of densely threaded fabric, it is worth noting that thicker areas and neps can occasionally contribute to increased friction between the threads.
The three primary problems related to yarns during this process are weft tears, warp ruptures, and yarn explosions. Whenever any of these difficulties arise, the machine must be stopped to allow the operator to reconnect the broken yarns before production may resume. Industrial output interruptions are a critical concern in manufacturing because they have an immediate influence on productivity, effectiveness, and profitability.
Harnessing the Power of Machine Learning for Quality Prediction
Benefits from applications like ML-based product quality prediction include reducing repair costs and shortening manufacturing lead times, as well as enhancing client relationships and having a better understanding of the root causes of issues. However, applying ML effectively is not a simple task. Data scientists must prepare the data (for instance, by encoding categorical characteristics), choose an ML method, and adjust its hyperparameters in order to produce meaningful machine learning models from the data.
The concept of AutoML, which involves automating the tasks that must be completed inside ML projects, offers a way to circumvent this resource shortage. By automatically generating wise judgments, AutoML lets individuals save precious resources, such as time, money, and human resources. Since effective machine learning algorithms and their hyperparameter settings are critical to the effectiveness of data learning, amalgamated algorithm selection and hyperparameter tuning become an important task in general AutoML systems as well as information processing pipelines.
Evaluating Automated Machine Learning Frameworks for Fabric Quality Prediction
This study aims to assist production planning using data acquired by IoT sensors and ERP systems from textile machines, as well as to produce a new quality forecast for each product based on imbalanced fabric quality data. In this approach, fewer people are needed for the quality control process. AutoML is used to streamline the ML training part and, as a consequence, shorten the data maintenance clutching operation process, as opposed to the traditional ML design by experts technique. It also stresses selecting the most suitable supervised ML method and optimizing the hyperparameters associated with it.
Comparative Analysis of Open-Source AutoML Tools
Seven contemporary open-source AutoML technologies are taken into account in the comparative study: FLAML, AutoViML, EvalML, AutoGluon, H2OAutoML, PyCaret, and TPOT. The literature’s present state as well as prior studies on AI-based quality control systems and textile forecast time improvement are covered in this part.
The evaluation of the proposed model’s predictive performance encompassed the utilization of four distinct metrics: the MAE, Explained Variance Score, mean squared error (MSE), and MAPE. The average absolute disparity among the observed values and the related predictions is represented by the MAE, which is computed over the whole dataset. This measure is used to assess the accuracy of predictions. Lower MAE scores signify higher predictive precision.
The findings are presented as the mean of the evaluation scores across ten external folds. The procedure involves identifying the optimal tools for each specific scenario by initially examining the average predicted score for each machine learning model and subsequently assessing the average computational effort, particularly in terms of training time.
Key Insights from the Comparative Analysis
The results reveal that EvalML emerges as the top-performing AutoML model for the predetermined objective function, particularly excelling in terms of mean absolute error (MAE). On the other hand, even with longer inference periods, AutoGluon performs better than other methods in measures like mean absolute percentage error (MAPE), root mean squared error (RMSE), and r-squared.
The main characteristics, as shown in Table 6, FabricRollMeter, EngineRpm, and LotNo have a higher influence on quality control. The performance of AutoGluon and EvalML is impressive, as shown in Fig. 2. EvalML achieves great efficiency by leveraging features, including PatternNo_cos and PatternNo_sin, that were not previously essential in feature importance, while the AutoGluon AutoML tool is identical to other tools in terms of attribute importance.
Balancing Predictive Accuracy and Computational Efficiency
The assessment of machine learning model performance is fundamentally contingent on model accuracy; nonetheless, the relentless pursuit of accuracy can precipitate an escalation in model complexity. One pragmatic approach to gauging such complexity entails the measurement of inference speed, typically referred to as ‘predict time.’
In our investigation, we have meticulously quantified ‘prediction duration’ while concurrently factoring in performance on the test data to comprehensively evaluate the constraints imposed by models generated through each respective framework. Conversely, we acknowledge that there exist contexts wherein inference time carries nominal significance. However, it is imperative to underscore that, within the purview of our study, prediction time assumes paramount importance.
The reason for this is that the new quality score will be predicted, and the loom will be adjusted according to this score. Despite the AutoGluon model garnering one of the highest scores, courtesy of its ensemble approach, it is concurrently associated with the longest prediction time.
Enhancing Model Interpretability and Leveraging Domain Expertise
Fabric samples are given quality labels based on the subjective assessments of human inspectors; as a result, the dataset produced may have biases or inconsistencies that skew the model’s predictions. In order to obtain a new quality attribute, we thus used the amount of errors, fabric length, and biased quality features through feature extraction.
Feature importance emerges as a salient facet of machine learning endeavors by affording practitioners the discernment of those attributes within a dataset that wield substantial influence over the ultimate prediction outcome, juxtaposed against features that bear comparatively diminished significance. Furthermore, this tableau attests to the efficacy of the sin/cos encoding technique, especially when tasked with enhancing the model’s capacity to discern categorical variables characterized by an extensive profusion of distinct, unique values.
Upon scrutinizing the AutoML tools showcased in Table 6, it becomes conspicuous that a substantial proportion of the sine and cosine features associated with the attribute attain prominent ranks in terms of feature importance. In textile manufacturing, the production process heavily relies on natural materials such as raw materials, yarn, and machine settings, each varying in quality and properties. Incorporating this variability into AutoML models, especially as new products are introduced, poses significant challenges requiring robust feature engineering and data preprocessing techniques.
Unlocking the Benefits of Automated Quality Control
By automating these processes with AutoML, companies significantly reduce labor costs associated with hiring and training personnel for inspection tasks. In addition, AutoML reveals patterns not immediately visible to operators and hidden correlations between production parameters and fabric quality. AutoML can detect potential quality issues early in the production process. It can detect quality problems not only at the level of finished fabric rolls but also at various stages of the production process.
By anticipating potential quality issues, manufacturers can implement preventative measures and process optimizations to minimize the likelihood of defects and maintain consistent quality standards. Early detection of quality outcomes results in significant cost savings by reducing material waste and rework. It detects deviations in fabric quality by analyzing production data resulting from factors such as loom features, pattern complexities, fabric differences, and engine power. It allows them to implement preventive measures and contingency plans to minimize the impact on quality standards. This results in higher-quality products that meet customer expectations and reduce the need for rework or returns.
Paving the Way for the Future of Textile Manufacturing
The idea of Industry 4.0 offers the chance to automatically gather ERP data as well as data from IoT sensors connected to industrial equipment. ML approaches can then be used to shorten lead times and increase productivity. Considering the latest state of technology today, it is a fact that the quality control process in the textile industry will become increasingly automated since machine learning can be applied to nearly any business.
In future studies, a recommendation system will be developed using the extracted domain information. This envisages a more efficient handling of yarn and fabric patterns, specifying the appropriate looms and engine rpm. With the intent of leveraging more open-source AutoML technology in the future, the goal involves expanding the dataset size. This expansion is particularly aimed at analyzing big data, where the application of deep learning has the potential to yield improved predictions.
Conclusion
This research employs machine learning tools for the training and evaluation of a regression task aimed at early-predicting fabric production quality, leveraging the capabilities of AutoML. To streamline the ML modeling endeavor, an array of AutoML tools, including FLAML, AutoViML, EvalML, AutoGluon, H2OAutoML, PyCaret, and TPOT, were methodically scrutinized. The benchmarking process encompassed an evaluation of computational effort and predictive performance.
The methodology for tool selection adopted a multi-faceted approach, wherein the best-performing tools were identified based on both the average predictive scores and the average computational effort expended for each specific task and scenario. For the selected objective function, EvalML offered the best average result among AutoML tools. The actual measurement values and the predicted values were found to have a strong correlation.
In real-world scenarios, the time it takes to make predictions (inference time) is a crucial factor. We conducted a comprehensive analysis of the trade-off between inference time and accuracy, uncovering notable disparities in the inference durations among the generated models. The most accurate frameworks achieve superior model precision, but at the expense of slower inference speeds.
As a consequence of these processes, the models obtained are often challenging for humans to comprehend, earning them the characterization of “black boxes.” Enhancing model interpretability, therefore, assumes paramount significance in enhancing the acceptability of AutoML outcomes among domain users. Within the ambit of this study, feature importance rankings were elucidated and subjected to comparison. Additionally, factors exerting an influence on quality estimation were shared and discussed in consultation with domain experts.
