Scholarlink

<p>Real-time monitoring is vital in modern manufacturing for ensuring product quality and minimizing downtime. However, conventional inspection methods in CNC lathe operations with live tooling often fall short in providing immediate quality feedback, impacting surface integrity and geometric accuracy. While vibration-based monitoring systems have shown promise, scalable and real-time solutions for assessing surface and geometric quality in live-tooling CNC lathe environments remain underexplored. This research presents a hybrid deep learning framework that integrates convolutional neural networks (CNN) and long short-term memory (LSTM) networks to predict surface roughness in milling and perpendicularity in drilling, within live tooling operations on a CNC lathe. High-frequency (50&#xa0;kHz) triaxial vibration signals are captured using an IEPE accelerometer mounted on a Haas ST-20 CNC lathe. The raw signals undergo multistage preprocessing, including noise filtering, time–frequency transformation, and Pearson correlation-based feature selection. The proposed model demonstrates strong predictive performance, achieving an accuracy of 99.57% for surface roughness (Ra) and 99.25% for perpendicularity. This non-contact, scalable framework enables real-time monitoring and minimizes reliance on post-machining inspection, thereby contributing to the advancement of Industry 4.0-compliant smart manufacturing systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of CNN-LSTM framework for predicting surface roughness and perpendicularity in live tooling operations via accelerometer sensor

  • 期刊论文

  • 发表日期：

    2025年10月18日

查看全文
  • Ashamoni Kakati,
  • Joseph C. Chen

摘要

Real-time monitoring is vital in modern manufacturing for ensuring product quality and minimizing downtime. However, conventional inspection methods in CNC lathe operations with live tooling often fall short in providing immediate quality feedback, impacting surface integrity and geometric accuracy. While vibration-based monitoring systems have shown promise, scalable and real-time solutions for assessing surface and geometric quality in live-tooling CNC lathe environments remain underexplored. This research presents a hybrid deep learning framework that integrates convolutional neural networks (CNN) and long short-term memory (LSTM) networks to predict surface roughness in milling and perpendicularity in drilling, within live tooling operations on a CNC lathe. High-frequency (50 kHz) triaxial vibration signals are captured using an IEPE accelerometer mounted on a Haas ST-20 CNC lathe. The raw signals undergo multistage preprocessing, including noise filtering, time–frequency transformation, and Pearson correlation-based feature selection. The proposed model demonstrates strong predictive performance, achieving an accuracy of 99.57% for surface roughness (Ra) and 99.25% for perpendicularity. This non-contact, scalable framework enables real-time monitoring and minimizes reliance on post-machining inspection, thereby contributing to the advancement of Industry 4.0-compliant smart manufacturing systems.