Industrial Manufacturing Quality Prediction Using an Edge Computing-based Framework

By Samudrapom DamReviewed by Susha Cheriyedath, M.Sc.Jan 23 2024

In an article recently published in the journal Scientific Reports, researchers proposed a synthetic minority oversampling technique (SMOTE)-extreme gradient boosting (XGBoost) quality prediction active control method based on edge computing to predict industrial product manufacturing quality.

Background

In recent decades, industrial big data has witnessed rapid development due to improved data processing and collection capabilities. Thus, the centralized model of big data processing based on cloud computing cannot effectively support industrial data analysis due to its several disadvantages, like difficulty in heterogeneous data integration from different sources, dealing with limited resources, and handling high broadband loads.

Addressing these issues is specifically crucial for equipment management systems that require real-time data processing, as any failure to detect faults in the equipment at the earliest opportunity can substantially diminish the product processing quality and result in significant losses throughout the industrial production line.

Recently, industrial-grade intelligent hardware-based edge computing technology has gained significant attention. It can establish a data bridge between cloud-based systems and production equipment, enabling rapid equipment operating status sensing in the industrial Internet of Things (IIoT) and intelligent adjustments. 

New-generation information technologies, such as artificial intelligence (AI) and big data, are rapidly integrating with the manufacturing industry with the emergence of intelligent manufacturing. One of their primary applications is to assist manufacturing facilities in predicting product quality.

Conventional predictive models primarily focus on establishing high-precision regression or classification models, with inadequate emphasis on imbalanced data. This scenario is common in real-world industrial environments concerning quality prediction.

The proposed approach

In this paper, researchers proposed a SMOTE-XGBoost quality prediction active control method based on joint optimization hyperparameters to address the issue of imbalanced data classification in product quality prediction. Additionally, edge computing technology was introduced to address the drawbacks of conventional cloud computing models, such as resource limitations and large bandwidth load, in industrial manufacturing.

The study's objective was to investigate an edge computing-based framework to predict the manufacturing quality of industrial products, guiding flexible industrial data handling. Initially, edge computing was introduced into product quality prediction to ensure higher reliability and shorter response times. Then, a method was designed to select quality-correlated parameters.

Eventually, the active control SMOTE-XGBoost method for quality prediction was used to address the imbalanced data classification problem within product quality prediction. The historical data-based proactive control method for quality prediction comprised two components: proactive control and quality prediction.

Initially, indirect dynamic process data was collected from production equipment, and crucial quality-related parameters were computed using mutual information. Then, these parameters were selected depending on their importance, which was followed by the division of the dataset into testing and training sets using stratified sampling.

Subsequently, the SMOTE algorithm obtained a balanced dataset, which was fed into the XGBoost-based predictive model for quality classification. A grid search method was also applied for the joint optimization of SMOTE and XGboost hyperparameters. Eventually, the optimal quality prediction model/SMOTE-XGboost_t was obtained and used for product quality prediction.

Experimental evaluation and findings

The effectiveness and practicality of the proposed method were evaluated through a case study of the brake disc production line. Analysis and selection of equipment process parameters based on the proposed active control method for quality prediction were performed for the brake disc production line using quality-correlated parameter selection to offer reference and guidance for the actual processing and production of brake disc products.

Four quality features, including clamping force, spindle current, spindle power, and spindle speed and feed speed, were selected to construct the prediction model. The selection of quality features played a crucial role in analyzing and predicting quality issues.

The coefficient of determination and the area under the curve (AUC) were used as evaluation metrics in comparative experiments performed to evaluate the reliability of predictive models, including SMOTE-logistic regression (LR), SMOTE-decision tree (DT), SMOTE-random forest (RF), SMOTE-support vector machine (SVM), and the optimized SMOTE-XGboost_t/proposed method.

Results from the comparative analysis of classification algorithms demonstrated that the proposed SMOTE-XGboost_t method had slightly higher AUC and coefficient of determination values compared to other classifiers while using the same SMOTE method. Specifically, the highest AUC value of the proposed method was 0.916, which indicated the feasibility of using the SMOTE-XGboost_t method to identify unqualified products effectively. ​​​​​​​Thus, the method could better predict the brake disc quality. Moreover, the SMOTE-XGBoost method with jointly optimized hyperparameters significantly improved the classification performance.

To summarize, the findings of this study demonstrated the effectiveness of the proposed method in addressing imbalanced data classification issues in practical industrial environments concerning quality prediction.