Application of Deep Learning Networks to Design Quality Control Process in the Motor Oil Industry

Abstract

Introduction: In light of recent advancements in the modern world, multivariate-multistage quality control patterns are increasingly recognized as vital and indispensable in manufacturing industries. This study delves into the significance and necessity of multivariate-multistage quality control in manufacturing, specifically focusing on motor oil production. As a foundational factor, motor oil quality considerably influences engine performance, lifespan, customer satisfaction, and market positioning. This research employs deep learning algorithms for monitoring and fault detection in quality components. The primary rationale for opting for deep learning algorithms over conventional statistical methods is the non-normal distribution of data and the large sample sizes, which can lead to inaccurate estimations and unstable analyses. Conversely, the unique capabilities of deep learning algorithms in handling complex data and extracting meaningful features from extensive motor oil production data justify their selection. Methods: To bolster accuracy and effective quality control, a combination of deep learning algorithms is utilized, including Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNN), and hybrid models such as LSTM-CNN, as well as Residual Networks (ResNet) with Dense Networks (DenseNet). The LSTM-CNN algorithm is applied to control numerical quality variables and identify temporal and sequential patterns in the data. Meanwhile, ResNet-DenseNet manages and analyzes visual data with non-uniform and intricate distributions. By integrating LSTM networks, CNNs, and residual connections, these algorithms excel at extracting meaningful features and capturing complex relationships within the data. This enhances performance and efficiency in quality control processes and facilitates intelligent decision-making. Such an approach is adept at uncovering latent patterns and intricate relationships between variables and quality attributes, enhancing quality control procedures and intelligent decision-making. Results and discussion: The amalgamation of these algorithmic capabilities enhances the efficacy of quality control processes, outperforming single-algorithm approaches. Additionally, the Bee Colony Clonal Algorithm (BCC) is employed to fine-tune the parameters of the LSTM-CNN and ResNet-DenseNet algorithms. This hybrid approach harnesses the Artificial Bee Colony (ABC) and Genetic Algorithm (GA) strengths, markedly improving the performance of deep learning algorithms in quality control and reducing the time required to achieve desired outcomes. To illustrate the practical applicability of the proposed algorithms, a case study in the motor oil production industry is examined. The proposed LSTM-CNN hybrid algorithm in fault detection demonstrated superior results compared to standalone CNN and LSTM algorithms, achieving performance improvements of approximately 15% and 8%, respectively. Furthermore, the proposed ResNet-DenseNet hybrid algorithm exhibited higher accuracy in visual components, enhancing performance by approximately 10% and 15% compared to ResNet and DenseNet algorithms, respectively. Conclusions: From both academic and practical standpoints, this research scrutinizes deep learning algorithms' influence on enhancing motor oil quality and efficiency. Advanced data analysis methods, particularly hybrid deep learning algorithms, are employed to identify quality patterns in production data.

Keywords

• auto-encoder
• deep learning
• fault detection
• metaheuristic algorithm
• quality control