Cloud-Based Machine Learning Methods for Parameter Prediction in Textile Manufacturing

Ray-I Chang, Jia-Ying Lin, Yu-Hsin Hung

TL;DR

This paper introduces a novel "query-based learning" approach for predicting product quality in textile manufacturing, outperforming traditional regression methods. By enabling dynamic interaction with the solution space, this method significantly reduces prediction errors, paving the way for more efficient and predictive maintenance in the industry.

What Problem Does This Paper Address?

Traditional textile manufacturing relies heavily on experienced technicians to manually adjust production parameters, a process that is time-consuming, resource-intensive, and prone to human error. This trial-and-error approach leads to inefficiencies and inconsistent product quality. The paper addresses the challenge of automating and optimizing this parameter adjustment process by developing a machine learning model that can accurately predict product quality based on manufacturing data. This is particularly important in the midstream textile manufacturing sector, which involves complex processes like warping, sizing, beaming, and weaving, where subtle parameter variations can significantly impact the final product.

Methods

The core of this paper is the development of a query-based regression model. Unlike traditional, unidirectional machine learning models, this approach incorporates a bi-directional interaction with the solution space. The model dynamically adjusts its learning process by:

  • Querying an "oracle": This involves selecting new data points (queries) that are near the current decision boundary.
  • Receiving feedback: The oracle provides the correct classification or value for these queried points.
  • Retraining: The model then uses this new information to refine its predictions and improve its understanding of the data.

This query-based learning is contrasted with preliminary models like Linear Regression, LASSO, Ridge Regression, and Elastic Net Regression. The query-based approach aims to overcome the limitations of these methods, such as underfitting or overfitting, by actively exploring the solution space and incorporating new, informative data points. The trained model is then deployed as a RESTful API for cloud-based analytics, enabling integration with other systems and facilitating automated notifications.

Datasets

The study utilized a dataset collected from the ERP system of a midstream textile manufacturer. This dataset comprised five sub-datasets related to:

  • Warping (37 columns, 26,103 records)
  • Sizing (46 columns, 15,950 records)
  • Beaming (31 columns, 37,089 records)
  • Weaving (25 columns, 161,821 records)
  • Inspection (12 columns, 481,610 records)

The data underwent extensive preprocessing, including:

  • Missing value imputation using circular interpolation.
  • Data transformation to split aggregated attributes.
  • Outlier removal (beyond three standard deviations).
  • Normalization using min-max scaling.

The preprocessed dataset was split into 60% for training and 40% for testing. The study also employed 10-fold cross-validation for robust model evaluation.

Key Insights and Ideas

  • Interactive Learning is Key: The paper strongly suggests that actively querying and learning from the data, rather than passively processing it, leads to significantly better predictive performance.
  • Targeted Data Exploration: By focusing on data points near the decision boundary, the query-based approach efficiently identifies areas where the model is uncertain, leading to more impactful learning.
  • Bridging the Gap: The proposed API and notification macro effectively bridge the gap between raw data analytics and actionable insights for production staff, enabling proactive maintenance.
  • Beyond Traditional Regression: While standard regression techniques provide a baseline, the dynamic nature of query-based learning offers a substantial improvement for complex, real-world prediction tasks.

Key Results

The primary quantitative result is the reduction in Mean Squared Error (MSE). The proposed query-based regression model achieved an average MSE of 0.0153, which is a significant improvement over the average MSE of the preliminary regression models (Linear, LASSO, Ridge, Elastic Net), which was 0.02. This represents an improvement rate of 22%.

Qualitatively, the study demonstrated that:

  • The query-based model, particularly when guided by retraining on samples with large prediction errors, outperformed random selection and learning from samples with small prediction errors.
  • The deployed API and automated notification system can effectively alert users to potential anomalies, facilitating timely intervention.

Limitations and Future Work

Limitations:

  • Dataset Specificity: The study was conducted on a midstream textile manufacturer's data. The performance of the model might vary with different types of textile products or manufacturing scenarios.
  • Attribute Limitations: The size and specific attributes of the industrial datasets used in this study might limit its generalizability.

Future Work:

  • Broader Data Analysis: Analyze data from a wider range of textile manufacturing processes, including computerized machines and manufacturing execution systems.
  • Correlation Analysis: Investigate the correlation between product quality and various manufacturing properties.
  • Enhanced Security: Implement single-sign-on for API access and encrypt data for enhanced privacy protection.
  • Attack Mitigation: Develop methods to address incorrect data injection attacks that could influence prediction results.
  • Industrial Validation: Validate the developed approach in industrial case studies and smart maintenance scenarios to demonstrate its efficiency and capacity.

Summary updated

Abstract

In traditional textile manufacturing, downstream manufacturers use raw materials, such as Nylon and cotton yarns, to produce textile products. The manufacturing process involves warping, sizing, beaming, weaving, and inspection. Staff members typically use a trial-and-error approach to adjust the appropriate production parameters in the manufacturing process, which can be time consuming and a waste of resources. To enhance the efficiency and effectiveness of textile manufacturing economically, this study proposes a query-based learning method in regression analytics using existing manufacturing data. Query-based learning allows the model training to evolve its decision-making process through dynamic interactions with its solution space. In this study, predefined target parameters of quality factors were first used to validate the training results and create new training patterns. These new patterns were then imported into the solution space of the training model. In predicting product quality, the results show that the proposed query-based regression algorithm has a mean squared error of 0.0153, which is better than those of the original regression-related methods (Avg. mean squared error = 0.020). The trained model was deployed as an application programing interface (API) for cloud-based analytics and an extensive auto-notification service.
BibTeX
@inproceedings{chang2024cloud,
  author = {Ray-I Chang and Jia-Ying Lin and Yu-Hsin Hung},
  title = {{Cloud-Based Machine Learning Methods for Parameter Prediction in Textile Manufacturing}},
  booktitle = {Sensors},
  year = {2024},
  url = {https://openreview.net/forum?id=QT16b0o6JO},
}

Loading related content...

Figure