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

Document Type : Original Article

Authors

1 Ph.D.Candidate in Industrial Engineering, Department of Industrial Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran.

2 Associate Professor,Department of Industrial Engineering ,Qazvin Branch, Islamic Azad University, Qazvin, Iran.

3 Associate Professor, Department of Industrial Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran.

10.48308/jimp.2023.232216.1462

Abstract

The utilization of multivariate-multistage quality control patterns is a crucial necessity in continuous manufacturing industries.context of continuous production processes, this research highlights the significant impact of incorporating multivariate control strategies to enhance the trajectory of final product quality. Integrating deep learning algorithms into the framework of multivariate-multistage quality control methods in the motor oil industry results in noteworthy benefits, including cost reduction in quality control, shortened production cycle times, decreased manufacturing expenses, and heightened customer satisfaction. This enhanced satisfaction can be attributed to the reduction in failures of engine components in vehicles. This paper introduces a substantial innovation by leveraging concatenated LSTM-CNN autoencoder algorithms for fault detection in numerical data and utilizing ResNet-DenseNet algorithms for analyzing image-based data. The optimization of model parameters is realized through the application of the GBC metaheuristic algorithm. Collectively, these innovations significantly improve Fault detection and control processes within the scope of this research. A practical case study of this research has been conducted at "Almoot," a motor oil manufacturing company. In summary, the proposed LSTM-CNN hybrid algorithm demonstrates a 15% improvement over the CNN algorithm and an 8% improvement over the LSTM algorithm in error detection processes. In the task of identifying Fault types, the LSTM-CNN hybrid algorithm outperforms the CNN algorithm by 8% and the LSTM algorithm by 10%. In the realm of visual components, the composite ResNet-DenseNet algorithm showcases a 10% enhancement compared to the standalone ResNet approach and a 4% improvement compared to the individual utilization of the DensNet algorithm.

Keywords

Main Subjects



Articles in Press, Accepted Manuscript
Available Online from 21 November 2023
  • Receive Date: 01 July 2023
  • Revise Date: 14 November 2023
  • Accept Date: 08 November 2023
  • First Publish Date: 21 November 2023
  • Publish Date: 21 November 2023