Assessment of Tool Wear With Insufficient and Unbalanced Data Using Improved Conditional Generative Adversarial Net and High-Quality Optimization Algorithm

The reliability of manufacturing tooling is key for intelligent manufacturing process, which requires accurately online identification of abnormal tool condition. However, in practical applications, insufficient and unbalanced data bring great difficulty for reliable tool wear assessment. In this study, a combined improved conditional generative adversarial net with high-quality optimization algorithm (CGAN-HQOA) is proposed to generate tool data having higher similarity with real data. Assessment of tool wear condition utilizing this newly generated data within convolutional neural network is shown with increased accuracy. The generated data can maintain sample diversity while minimizing deviation from real sample characteristics with CGAN-HQOA. The effectiveness is investigated using unbalanced data under various scenarios, where the quality of generated data from the proposed model is compared to those from commonly used data generation algorithms, such as generative adversarial nets. Moreover, the robustness of the proposed method is investigated by using different cutting tools. Results demonstrate that with the proposed model, better quality data can be generated, and more accurate tool wear condition can be assessed using generated data. The findings will be beneficial in practical applications where only limited test data are available, whereas accurate and online tool wear can be evaluated with proposed method.

Automated summary

The reliability of manufacturing tooling is crucial for intelligent manufacturing processes. However, limited and unbalanced data pose challenges for accurate tool wear assessment. This study proposes a combined CGAN-HQOA model that generates tool data with higher similarity to real data, resulting in improved tool wear condition assessment using convolutional neural network. The effectiveness of the proposed method is verified using unbalanced data and different cutting tools, demonstrating the superiority of the generated data and the accuracy of tool wear assessment. These findings are valuable for practical applications with limited test data.